Draft version January 28, 2019

Preprint typeset using LATEX style AASTeX6 v. 1.0

HOW TURBULENCE ENABLES CORE-COLLAPSE SUPERNOVA EXPLOSIONS

Quintin A. Mabanta1 and Jeremiah W. Murphy2

1
Florida State University, qam13b@my.fsu.edu
2
Florida State University, jwmurphy@fsu.edu

ABSTRACT
arXiv:1706.00072v2 [astro-ph.HE] 24 Jan 2019

An important result in core-collapse supernova (CCSN) theory is that spherically-symmetric, one-

dimensional simulations routinely fail to explode, yet multi-dimensional simulations often explode.
Numerical investigations suggest that turbulence eases the condition for explosion, but how is not
fully understood. We develop a turbulence model for neutrino-driven convection, and show that this
turbulence model reduces the condition for explosions by about 30%, in concordance with multi-
dimensional simulations. In addition, we identify which turbulent terms enable explosions. Contrary
to prior suggestions, turbulent ram pressure is not the dominant factor in reducing the condition
for explosion. Instead, there are many contributing factors, ram pressure being only one of them,
but the dominant factor is turbulent dissipation (TD). Primarily, TD provides extra heating, adding
significant thermal pressure, and reducing the condition for explosion. The source of this TD power is
turbulent kinetic energy, which ultimately derives its energy from the higher potential of an unstable
convective profile. Investigating a turbulence model in conjunction with an explosion condition enables
insight that is difficult to glean from merely analyzing complex multi-dimensional simulations. An
explosion condition presents a clear diagnostic to explain why stars explode, and the turbulence model
allows us to explore how turbulence enables explosion. Though we find that turbulent dissipation is
a significant contributor to successful supernova explosions, it is important to note that this work is
to some extent qualitative. Therefore, we suggest ways to further verify and validate our predictions
with multi-dimensional simulations.
Keywords: supernovae: general — hydrodynamics — methods:analytical — methods: numerical —
shock waves — turbulence

1. INTRODUCTION multi-dimensional simulations do explode (Herant et al.

Understanding how massive stars end their lives still 1994; Janka & Müller 1995; Burrows et al. 1995; Janka
remains an important astrophysical problem. Observa- & Müller 1996; Burrows et al. 2007; Melson et al. 2015;
tions indicate that most massive stars likely explode Dolence et al. 2015; Müller 2016; Roberts et al. 2016;
as core-collapse supernovae (Li et al. 2011; Horiuchi Bruenn et al. 2016) and initial analyses suggest that
et al. 2011), but the theoretical details of the explo- multi-dimensional instabilities and turbulence aid the
sion mechanism are still uncertain. The most certain neutrino mechanism toward explosion (Murphy & Bur-
part of the theory is that the Fe core collapses, bounces rows 2008; Marek & Janka 2009; Murphy & Meakin
at nuclear densities forming a proto-neutron star, and 2011a; Hanke et al. 2012; Murphy et al. 2013; Radice
launches a shock wave (Janka et al. 2016). However, et al. 2016; Couch & Ott 2015; Melson et al. 2015).
this shock wave quickly stalls (Hillebrandt & Mueller Therefore, understanding the explosion mechanism re-
1981; Mazurek 1982; Mazurek et al. 1982). For over quires an understanding of the conditions between failed
three decades, the most prominent mechanism has been and successful explosions and how turbulence aids the
the delayed-neutrino mechanism in which neutrinos re- explosion. There appears to be a critical condition for
heat the matter below the stalled shock and relaunch it explosion (Burrows & Goshy 1993; Murphy & Burrows
into an explosive blast wave (Bethe & Wilson 1985). For 2008; Murphy & Dolence 2017), and the critical condi-
all but the least massive stars, the neutrino mechanism tion for explosion is easier to obtain in multi-dimensional
fails in one-dimensional, spherically-symmetric simula- simulations (Herant et al. 1994; Janka & Müller 1995,
tions (Liebendörfer et al. 2001b,a, 2005; Rampp & Janka 1996; Burrows et al. 1995; Murphy & Burrows 2008;
2002; Buras et al. 2003; Thompson et al. 2003; Kitaura Melson et al. 2015). In this manuscript, we propose an
et al. 2006; Buras et al. 2006a; Radice et al. 2017) but analytic model for turbulence and investigate how it re-
2

duces the condition for explosion. dimensional simulations seem to succeed (sometimes
For a thorough review on the status and problems weakly) (Herant et al. 1994; Janka & Müller 1995; Bur-
of CCSN theory, see Müller (2017) and Janka et al. rows et al. 1995; Janka & Müller 1996; Burrows et al.
(2016). Here, we motivate our work with some of the 2007; Melson et al. 2015; Dolence et al. 2015; Müller
most salient points. We know that the core collapses, 2016; Roberts et al. 2016; Bruenn et al. 2016; Burrows
but not yet how the collapse reverses into explosion. et al. 2016), and the best indications are that turbu-
Whether a massive star explodes or not, the Fe core lence plays an important role in aiding the delayed-
collapses at the end of the massive star’s life. Prior to neutrino mechanism toward explosion (Bethe & Wilson
collapse, the overlying Si-burning layer adds Fe onto the 1985; Janka & Müller 1996; Marek & Janka 2009; Mur-
iron core. As the iron core grows in size, the electrons phy et al. 2013; Murphy & Meakin 2011a). Therefore,
which supply much of the electron degeneracy pressure to truly understand the explosion mechanism of massive
become more and more relativistic. As the core nears stars, we need to identify the conditions for explosion
the Chandrasekhar mass limit, the relativistic electron and how turbulence affects these conditions.
degeneracy pressure becomes less effective at support- There are many attempts to characterize these con-
ing the core against gravitational collapse. Meanwhile, ditions, some are empirical (O’Connor & Ott 2011; Ott
neutrino losses reduce the lepton number, decreasing the et al. 2013; Ertl et al. 2016), some are heuristic (Janka
number of electrons available to supply pressure. The & Keil 1998; Thompson 2000; Thompson et al. 2005;
iron core becomes gravitationally unstable and contracts Buras et al. 2006b; Müller et al. 2016), and others at-
down to a sea of nucleons, forming a proto-neutron star, tempt to derive a condition from fundamentals (Burrows
and is thus supported mostly by the strong force. At & Goshy 1993; Pejcha & Thompson 2012; Murphy &
these nuclear densities, the equation of state for the core Dolence 2017). Of these, the most illuminating to date
stiffens, and the core abruptly bounces, slamming into has been the neutrino-luminosity and accretion-rate crit-
the rest of the star which is collapsing supersonically ical curve (Burrows & Goshy 1993). During the stalled
onto the bouncing proto-neutron star. This creates an shock phase, Burrows & Goshy (1993) noted that one
outward moving shock wave. As the shock wave prop- may derive steady-state solutions for the stalled shock
agates outward, it loses energy via photodissociation of structure. The governing equations describe a bound-
Fe, electron capture, and neutrino losses. The shock ary value problem in which the lower boundary is set
stalls into an accretion shock, but if the star is to ex- by the neutron star surface (the neutrino-sphere) and
plode, this stalled accretion shock must relaunch into an the outer boundary is the shock. They parameterized
explosive blast wave. the problem in terms of the accretion rate, Ṁ, onto
One proposed solution to relaunching the blast wave the shock, and the neutrino luminosity, Lν , emanat-
was the delayed-neutrino mechanism (Bethe & Wilson ing from the core. In this two-dimensional parameter
1985). During the stalled shock phase, the neutron star space they found steady-state solutions below a critical
is cooling with a neutrino luminosity of a few × 1052 curve. Above this curve, they did not find steady-state
ergs/s, but only about 10% of this luminosity is recap- stalled solutions, and suggested, but did not prove, that
tured in a net heating region, the gain region. This the solutions above the critical neutrino-luminosity and
volume is above the proto-neutron star but below the accretion-rate curve are dynamic and explosive.
shock. If neutrino heating were the only effect, then this Murphy & Dolence (2017) take a step closer to prov-
would be sufficient to relaunch the explosion. However, ing that the solutions above the curve are explosive by
the region below the shock is in sonic contact, and so the showing that the only steady solutions above this curve
structure satisfies a boundary-value problem. While the indeed have a positive shock velocity (vs > 0). Burrows
neutrino heating adds heat that would drive explosion, & Goshy (1993) focused on just Ṁ and Lν , but there are
the neutrino cooling at the base and the ram pressure five parameters that define the steady solutions. They
of matter accreting through the shock keeps the shock are neutrino luminosity (Lν ), mass accretion rate (Ṁ),
stalled. The hope has been that for high enough neu- neutron star mass (MN S ), neutron star radius (RN S ),
trino luminosities or low accretion rates, the neutrinos and the neutrino temperature (Tν ). Murphy & Dolence
may overwhelm the ram pressure and relaunch the ex- (2017) point out that the critical condition is not a crit-
plosion. Alas, spherically symmetric simulations show ical curve but a critical hypersurface; most importantly,
that this mechanism fails in in all but the least massive they find that this critical hypersurface is described by
progenitors (Liebendörfer et al. 2001b,a, 2005; Rampp one dimensionless parameter, Ψ. In essence, Ψ is an in-
& Janka 2002; Buras et al. 2003; Thompson et al. 2003; tegral condition related to the balance of pressure and
Kitaura et al. 2006; Buras et al. 2006a; Müller et al. gravity behind the shock. For a given set of the five
2017; Radice et al. 2017). parameters, Ψ may be negative, zero, or positive, which
While one-dimensional simulations fail, many multi- correspond to vs < 0, vs = 0, and vs > 0. There is
 3

always a minimum Ψ, and if Ψmin < 0, then there is dwell-time distributions actually lead to enhanced en-
always a stable steady-state stalled solution such that tropy in the gain region. Couch & Ott (2015) revis-
Ψ = 0. The critical condition is where Ψmin = 0. Above ited the idea of turbulent ram pressure being the main
this critical condition, Ψmin > 0, and all steady solutions multi-dimensional contribution, but speculated it as an
have vs > 0. Assuming that these vs > 0 steady solu- effective ram pressure, not distinguishing between multi-
tions correspond to explosion, they use Ψmin to define dimensional effects. At this point many of these sugges-
an explodability parameter. tions seem plausible, but it is not clear which, if any,
Using one- and two-dimensional simulations Murphy explain why turbulence aids explosions. In fact, we will
& Burrows (2008), empirically confirmed that critical- show that none of these explanations truly captured the
ity is a useful condition for explosion in core-collapse role of turbulence. However, we do note that the higher
simulations. Furthermore, they found that the critical entropy profiles of multi-dimensional turbulence should
condition is ∼30% lower in two-dimensional simulations. have hinted that turbulent dissipation plays an impor-
Subsequently, others have confirmed these findings and tant role.
that the reduction is similar in three-dimensional sim- Part of the reason it was difficult to assess how
ulations (Fernández 2015a; Hanke et al. 2012; Dolence turbulence aids explosion is the complexity of multi-
et al. 2013; Handy et al. 2014). dimensional simulations. In these large, non-linear sim-
Initial indications are that turbulence causes this re- ulations it is difficult to isolate the causes and effects
duction but these investigations were mostly suggestive of turbulence. In this paper, we propose a different,
and not conclusive. Decades ago, Bethe (1990) recog- more illuminating approach. We model turbulence in
nized the potential importance of neutrino-driven con- the context of the critical condition. Because we have
vection aiding explosion. This initial investigation sug- direct control over how turbulence affects the equations,
gested that turbulent ram pressure behind the material we can directly assess the causes and effects of our tur-
would push against the shock. In the early 90s, the first bulence model in reducing the critical condition.
two-dimensional simulations with crude neutrino trans- To do this, we extend the explodability parameter of
port exploded while the one-dimensional simulations did Murphy & Dolence (2017) to multiple dimensions by
not. These investigators speculated that neutrino-driven including a turbulence model. Currently, a turbulence
convection aided the explosion (Burrows et al. 1995; model exists for neutrino driven convection (Murphy
Janka & Müller 1996; Janka 2001; Colgate & Herant & Meakin 2011a; Murphy et al. 2013), but one does
2004). not yet exist for the SASI. Thus, the only viable av-
Blondin et al. (2003) identified a new instability that enue of an analytic investigation of turbulence is through
can also drive turbulence: the standing accretion shock neutrino-driven convection. Investigations on how tur-
instability (SASI). Linear analyses suggest that this bulence driven by the SASI affects the condition will
instability results from an advective-acoustic feedback have to wait until we have a valid turbulence model for
cycle (Foglizzo & Tagger 2000; Foglizzo et al. 2006; the SASI. Therefore, we use the neutrino-driven convec-
Sato et al. 2009a,b; Guilet et al. 2010), and subse- tion turbulence model of Murphy et al. (2013) in the
quent investigations considered the possibility that the critical condition of Murphy & Dolence (2017).
SASI aids the delayed neutrino mechanism toward ex- In section 2, we revisit criticality, present the neutrino-
plosion (Marek & Janka 2009; Hanke et al. 2012, 2013; driven turbulence model, derive the equations, and de-
Fernández et al. 2014). Instead, Murphy et al. (2013) scribe the solution method. Furthermore, we derive an
found that in simplified simulations convection dom- analytic upper limit on the Reynolds stress. In section 3
inates just before and during neutrino-driven explo- we present how turbulence modifies the structure of the
sions. Further analyses with less simplistic neutrino ap- post-shock flow and how it affects the critical condition.
proaches found that the SASI does dominate at times, Finally, in section 4, we conclude and discuss implica-
but convection likely dominates for most but not all ex- tions for the core-collapse mechanism and future inves-
plosion conditions (Müller 2016; Radice et al. 2016; Bur- tigations. Our main conclusions are that turbulent ram
rows et al. 2012; Murphy et al. 2013; Murphy & Meakin pressure is not the primary turbulent term aiding explo-
2011a). sions, rather it is one of a few and the dominant turbu-
In one attempt to find the reason why turbulence aids lent effect aiding explosions is turbulent dissipation.
explosion, Murphy & Burrows (2008) found that entropy
is higher in the multi-dimensional case. At the time, 2. METHODS
they as well as others suggested that a longer dwell time In this section, we outline the method for deriving
in the gain region causes this higher entropy (Thompson the critical condition including the effects of neutrino-
et al. 2005; Marek & Janka 2009; Buras et al. 2006b), driven convection. Our primary goal is to incorporate a
however, these investigations did not show that these turbulence model in the integral explosion condition of
4

Murphy & Dolence (2017) which is a generalization of at the neutrinosphere with the stalled shock solution.
the foundational work of Burrows & Goshy (1993). They suggested, but did not prove, that the solutions
Our attempt is not the first to include turbulence above the Lν − Ṁ curve are explosive.
in calculations of critical conditions. Yamasaki & Ya- Murphy & Dolence (2017) took one step closer in prov-
mada (2005) and Yamasaki & Yamada (2006) consid- ing that the solutions above the critical curve are explo-
ered several effects that might reduce the explosion sive. They showed that the only steady solutions above
condition, including rotation and convection. How- the curve have vs > 0, lending support to the suppo-
ever, their attempt to model turbulence only included sition of Burrows & Goshy (1993). They did this by
a term representing turbulent enthalpy flux. A more connecting the discriminant of Burrows & Goshy (1993)
self-consistent derivation of a turbulence model should (the τ = 2/3 condition) to a new dimensionless param-
include three terms at a minimum: the turbulent en- eter Ψ. This Ψ parameter is a measure of overpressure
thalpy flux, the turbulent ram pressure, and turbulent compared to hydrostatic equilibrium and directly indi-
dissipation. Without including all of these effects, it is cates whether the shock would move out, in, or stay sta-
unclear how turbulence would actually effect the critical tionary. Because it is more straightforward, we do our
condition for explosion. In this manuscript, we incorpo- calculations in the Burrows & Goshy (1993) formalism,
rate the turbulence model of Murphy et al. (2013) which using the neutrino density discriminant, but we will also
includes these three terms and has been validated with present our results in the context of Ψ, the overpressure.
core-collapse simulations. While the method of Burrows & Goshy (1993) is sim-
First, in subsection 2.1, we revisit criticality and pler, Murphy & Dolence (2017) provides a more direct
present the relevant equations and assumptions. Next, connection to vs , and thus explosion.
in section 2.2, we present our methods for incorporating The first step in finding the critical curve is find-
turbulence into these equations. To find the solutions ing steady-state solutions to the Euler equations. The
for both the average background flow and the turbulent steady state equations are
flow, we decompose the equations into the average and
∇ · (ρu) = 0 , (1)
turbulent quantities, commonly called Reynolds decom-
position. By decomposing the variables into average and
turbulent flows, we introduce extra variables. To close ∇ · (ρu ⊗ u) = −∇P − ρ∇Φ , (2)
the system of equations, we motivate a turbulence model
in 2.3, and we specify our assumptions. In 2.4 we de- and
u2
  
scribe our method for solving this system of equations.
∇ · ρu h + = −ρu · ∇Φ + ρq . (3)
As it turns out, there is maximum allowable Reynolds 2
stress in the stalled shock solutions. Even though this Where ρ is mass density, u is velocity, P is pressure, Φ is
upper limit does not affect the critical condition, we de- gravitational potential, h is enthalpy, and q is the total
rive an analytic expression for it in section 2.5 and in- heating. In general, heating and cooling by neutrinos is
clude it in our critical condition calculations. Finally, in best described by neutrino transport (Janka 2017; Tam-
section 2.6, we discuss some of our assumptions and pa- borra et al. 2017); we simplify neutrino transport by
rameters that need to be verified by multi-dimensional invoking a simple light-bulb prescription for neutrino
simulations heating and a local cooling (Janka 2001)
2.1. The Critical Curve  6
Lν κ T
q= − C0 . (4)
Two explosion conditions which start from first prin- 4πr2 2M eV
ciples are the critical curve of Burrows & Goshy (1993) I.e., we’ve adopted a spherically symmetric neutrino
and a generalization of this condition, the Ψ condition source. r is the radius from this source, Lν is the neu-
of Murphy & Dolence (2017). We will modify these crit- trino luminosity emitted from the core of the star, and
ical conditions to include turbulence, so first we revisit
what constitutes a critical condition.  2
Burrows & Goshy (1993) introduced a critical curve Tν
κ = 7.5 × 10−28 (Yn + Yp )[g · cm2 ] (5)
that divides steady-state solutions from no solutions in 4M eV
the Lν − Ṁ plane. Below this curve, one can find steady- is the neutrino opacity (Murphy et al. 2013); where Yn
state stalled shock solutions that satisfy all boundary and Yp are the neutrino and proton fractions per baryon.
conditions. Above the curve, there are no stalled-shock κ is related to the optical depth by
solutions which satisfy all of the boundary conditions.
Zrs
In detail, the main discriminant for finding steady-state
τ= ρκdr , (6)
solutions is whether or not they could match the density
rg
 5

which dictates the absorptivity of the material in the 2.2. Reynolds Decomposed Equations
gain region; rs and rg are the positions of the shock and A standard method to incorporate turbulence is
gain radius, respectively. T is the matter temperature, through Reynolds decomposition. The primary goal is
and C0 is the cooling factor (1.399 × 1020 ergs/g/s). to derive mean-field, steady-state equations for turbu-
Following in the steps of Burrows & Goshy (1993) lence. The first step is to Reynolds decompose the vari-
equations (1-3) represent a boundary value problem with ables into background and perturbed components; i.e.
the boundaries being the neutron star surface and the u = u0 +u0 , where 0 denotes the background component
shock. At the shock, we want the pre-shock, inflowing and the prime indicates the turbulent term. Next, one
material to match the post-shock material through the substitutes these terms into the time-dependent Navier-
Rankine-Hugoniot jump conditions. For the moment, Stokes equations. To obtain the mean-field correla-
let us assume that vs is zero, and in this case, the jump tions for turbulence, one averages the equations both
conditions become: in time and solid angle. For simplicity, we denote both
ρ1 u1 = ρ2 u2 , (7) of these averages by the operator h·i. Since hui = u0 and
hu0 i = 0, terms that involve one component of a turbu-
lent variable are zero. All non-zero turbulent terms are
ρ1 u21 + P1 = ρ2 u22 + P2 , (8) higher order correlations of turbulent variables.
Technically, the turbulence represents a time-
and
    dependent fluctuation. However, the mean-field vari-
1 P1 1 2 P2 ables, or turbulent correlations are time-averaged cor-
ρ1 u1 e1 + u21 + = ρ2 u2 e2 + u2 + .
2 ρ1 2 ρ2 relations and can be in steady-state. For core-collapse
(9) simulations, the turbulent correlations are effectively in
Where e is the internal eneregy and the subscripts 1 and steady-state, so one may drop the time derivatives in
2 indicate the downstream and upstream flows at the the Reynolds-averaged equations. The resulting equa-
jump, respectively. Normally, one uses eq. (7) in eq. (9) tions represent steady-state equations for the back-
to eliminate the mass flux in the Hugoniot-Rankine ground flow and the mean-field turbulent correlations.
jump condition. Here, we explicitly include it because In this manuscript, we highlight the important correla-
this term can not be neglected when we Reynolds de- tions and steady-state equations. For a more thorough
compose and derive the jump conditions including the derivation of the equations from the Navier-Stokes equa-
turbulent terms. One then integrates inward to the neu- tions, see Meakin & Arnett (2007) or Murphy & Meakin
tron star surface where the density profile must match (2011b).
the neutron star surface such that the neutrino optical The three dominant Reynolds turbulent correlations
depth is 2/3. If the neutrino optical depth is not 2/3, are the Reynolds Stress (R), turbulent dissipation
then one searches for a new rs so that the shock and (k ), and turbulent luminosity (Le ) (Murphy & Meakin
neutron star boundary conditions are met. In practice, 2011a; Murphy et al. 2013). The Reynolds stress is the
Yamasaki & Yamada (2005) noticed that the density at turbulent fluctuation in momentum stress, turbulent lu-
the neutrinosphere has about the same value in most minosity is the transport of turbulent internal energy,
situations. For the opacities used in this manuscript, and turbulent dissipation is the viscous conversion of
Murphy & Burrows (2008) calculated the density in mechanical energy to heat. These terms are
one-dimensional and two-dimensional simulations to be
about 7 × 1010 g cm−3 at the neutron star “surface.” Rij = u0i u0j , (10)
Since this density condition is faster to integrate, we
use it instead of the neutrino optical depth condition.
While Burrows & Goshy (1993) provide an elegant Le,i = 4πr2 ρ0 hu0i e0 i = 4πr2 Fe,i , (11)
one-dimensional explosion condition, it does not accu- and in the limit of small viscosity, turbulent dissipa-
rately diagnose realistic supernova explosions in multiple tion is
dimension (Murphy & Burrows 2008). However, subse-  = 2ν(∇u0 ) · (∇u0 ) . (12)
quent multi-dimensional work suggests that one might
be able to augment the technique for finding the critical The turublent kinetic energy dissipation is k =
curve to include turbulence (Murphy et al. 2013; Mur- tr()/2. Note that this definition is slightly different
phy & Meakin 2011a; Hanke et al. 2012; Couch 2013; from the Murphy & Meakin (2011b) definition which
Radice et al. 2016). To do this, we build on the work presented a confusing sign and needlessly included tur-
of Murphy & Meakin (2011a) and Murphy et al. (2013), bulent diffusion in the turbulent dissipation term. Here,
and use the Reynolds decomposed continuity equations we take the limit of small viscosity so the turbulent diffu-
to find a multi-dimensional critical curve. sion term goes away, but turbulent dissipation remains.
6

Furthermore, we corrected the sign so that a positive ρ0 u0 hu02 /2i) is the kinetic energy luminosity. Now that
 corresponds to taking energy from the kinetic energy we have introduced three new turbulent variables, we
equation and putting it in the internal energy equation. have a total of six unknown variables and only three
Since this term requires higher order correlations, we equations, necessitating more equations.
model it using Kolmogorov’s assumptions (see section
2.3 or refer to the result of Canuto (1993) for a more 2.3. Turbulence Models
robust description). Including the turbulent components, the Reynolds-
The resulting steady-state, Reynolds-decomposed decomposed conservation equations, (13-15), now have
equations are more variables than equations. These extra variables are
the Reynolds stress, turbulent luminosity, and turbulent
∇ · (ρ0 ~u0 + hρ0 ~u0 i) = 0 , (13)
dissipation. Therefore, to find a solution to these equa-
tions, we need a turbulence closure model. Turbulence
hρ~ui · ∇~u0 = −∇P0 + ρ0~g − ∇ · hρRi , (14) depends upon the bulk macroscopic flow, so the equa-
tions for turbulence represent a boundary value prob-
and
lem that depends upon the specifics of the background
hρui · ∇e0 + hP0 ∇ · u0 i + hP 0 ∇ · u0 i = flow. For this reason, Murphy et al. (2013) developed a
(15)
−∇ · Fe + ρ0 q + ρ0 k . global turbulence model for neutrino-driven convection.
The steady-state equations require local turbulent ex-
To see the exact equation, please refer to Meakin & Ar-
pressions and derivatives. Therefore, to use the global
nett (2007). There, they have fully expanded the above
model, we must make some assumptions and translate
equation into their background and perturbed compo-
the global model to a local model.
nents. The internal energy flux and Reynolds stress are
In the core-collapse problem, there may be two sources
of turbulence: convection and the SASI (Bethe 1990;
Fe = hρue0 i (16a)
Blondin et al. 2003). In principle, to correctly model
and turbulence in the core-collapse problem, we need a tur-
hρRi = hρuu0 i . (16b) bulence model that addresses both driving mechanisms:
convection and the SASI. While there are nonlinear
Alternative and common definitions are Fe = ρ0 hu0 e0 i
models to describe turbulent convection, there are no
and hρRi = ρ0 hu0 u0 i. We use eq. (16a) because it
nonlinear models yet to describe SASI turbulence. Thus,
gives the same result but is much simpler and cleaner
we proceed with a convection based analysis of turbu-
to calculate in numerical simulations. Expanding equa-
lence.
tions (16a) and (16b) gives
There are five turbulent variables (R, Le , k ), three of
Fe = ρ0 hu0 e0 i + hρ0 u0 e0 i + u0 hρ0 e0 i (17a) them are Reynolds stress terms (Rrr , Rφφ , and Rθθ );
therefore, we invoke five constraints. Our five global
and
constraints are as follows. First, we eliminate the tan-
hρRi = ρ0 hu0 u0 i + hρ0 u0 u0 i + u0 hρ0 u0 i . (17b) gential components of the Reynolds stress. In neutrino-
driven convection, the radial Reynolds stress is in rough
Within the convective region, Murphy & Meakin
equipartition with both of the tangential components
(2011b) and Murphy et al. (2013) found that the first
(Murphy et al. 2013):
term is the dominant term. However, just using the
first term creates a large spike at the aspherical shock Rrr ∼ Rφφ + Rθθ . (20)
which has nothing to do with convection and everything
Similar simulations showed that the transverse compo-
to do with the jump conditions across the aspherical
nents are roughly the same scale:
shock. However, using eq. (16a) mitigates this problem
and gives the correct turbulent energy flux within the Rφφ ∼ Rθθ (21)
convective region.
Using Kolmogorov (1941)’s hypothesis we relate the
The Decomposed boundary conditions are:
Reynolds stress to the turbulent dissipation:
ρ1 u21 + P1 + ρ1 Rrr = ρ2 u22 + P2 (18) 3/2
u03 Rrr
k ≈ = , (22)
and L L
P1 Le Lk 1 2 P2 1 where L is the largest turbulent dissipation scale. From
+ e1 + + + u1 + Rrr = + e2 + u22 , (19)
ρ1 Ṁ Ṁ 2 ρ2 2 Murphy & Meakin (2011a), we note that buoyant driv-
ing roughly balances turbulent dissipation:
where Ṁ = 4πr2 ρu is the mass accretion rate, Le =
4πr2 ρ0 hu0 e0 i is the internal energy luminosity, and Lk (= Wb ≈ Ek , (23)
 7

where the buoyant driving is the total work done by bulent dissipation rate scales as the perturbed velocity
buoyant forces in the convective region, cubed over the characteristic length of the instabilities,
Zrs equation (22). Numerical simulations suggest that the
Wb = hρ0 u0i ig i dV , (24) largest dissipation scale in convection is the size of the
convective zone (Couch & O’Connor 2014; Foglizzo et al.
rg
2015; Fernández 2015b), or the gain region in the core-
and the total power of dissipated turbulent energy is collapse case. Moreover, eqs. (23-26) have a global defi-
Zrs nition of k , and so we assume that k is roughly constant
Ek = ρk dV . (25) over the gain region. Therefore, from equation (25),
Z rs
rg
Ek = ρk dV , (28)
Lastly, three-dimensional simulations from Murphy rg

et al. (2013) show that the source of neutrino-driven we have

convection, the neutrino power, is related to the turbu- Ek Wb
k ≈ = . (29)
lent dissipation and the turbulent luminosity by Mgain Mgain
Lν τ ≈ Ek + Lmax
e (26) The final missing piece is the Turbulent Luminosity,
Le , and we connect this to the turbulent dissipation by
Together, equations (20-23) and (26) represent our tur-
rewriting the buoyant work in terms of the turbulent
bulence closure model.
luminosity. We combine the observation that buoyant
Now that we have combined a series of global con-
work is approximately equal to the turbulent dissipation
ditions to close our global model, we must relate these
power (Murphy et al. 2013) and that this energy can be
back to local functions in order to incorporate them into
converted to heat (Murphy & Meakin 2011a). Ignoring
our conservation equations. We do this by making as-
compositional perturbations, the density perturbation
sumptions about the local profile for each term and scal-
in terms of the perturbed energy and pressure is
ing them by one parameter for each turbulent term. In
translating from global to local, we introduce three pa- ∂ρ
∂ρ

ρ0 = e0 P
+ P0 . (30)
rameters of the turbulent region: a constant Reynolds ∂e ∂P e
stress (R), a constant dissipation rate (k ), and a max- Convective flows are generally dominated by buoyant
imum for the turbulent luminosity (Lmax e ); the corre- perturbations and not pressure perturbations. Even for
sponding local terms are ∇ · hρRi, ρ0 k , and ∇ · hL ~ ei high mach number convection, buoyancy tends to dom-
respectively (see equations (14-15)). Thus, the final so- inate; see Murphy et al. (2013). Therefore, one may
lution for turbulence boils down to finding these three express the density perturbation in terms of the energy
parameters. perturbation alone. Applying the above step and some
To find the three parameters, we insert the profiles for algebra to (24), we obtain
the turbulent terms and their parameters into the global
(γ − 1)Le Φρ dr
Z
conditions, equations (20-23) and (26). This then leads Wb = . (31)
to a set of equations for the parameters, and we use sim- P r
ple algebra to solve for the scale of the parameters that Now that we have an equation for the buoyant driving
satisfies those global conditions. We solve for these in as a function of the turbulent luminosity (Le ), we now
the order they are presented: the Reynolds stress, tur- need the radial profile for Le to complete the connection
bulent dissipation, and finally the turbulent luminosity. between turbulent luminosity and buoyant driving. Ini-
Our first task is to reduce the three Reynolds stress tial numerical investigations of Le suggest that it rises
terms down to one. In neutrino-driven convection, there from zero at the gain radius to a nearly constant value
is a preferred direction (i.e. in the direction of grav- above the gain radius until the shock (Murphy et al.
ity) and simulations show that there is an equipartition 2013). Therefore, we develop an ansatz for Le which
between the radial direction and both of the tangen- roughly satisfies the shape seen in simulations
tial directions (Murphy et al. 2013). Simulations also  
max r − rg
show that the two tangential directions have the same Le = Le tanh . (32)
scale (Murphy et al. 2013). Evaluating these assump- h
tions in equations (20-21) reduces the representation of Where h is the distance it takes for the turbulent lumi-
the Reynolds stress as three variables down to one: nosity to increase from zero to roughly Lmax
e . Thus, our
buoyant driving power becomes
R = 2Rrr . (27)
 
r − rg Φρ dr
Z
Kolmogorov’s theory of turbulence predicts the tur- Wb = (γ − 1)Lmaxe tanh . (33)
h P r
8

Finally, relating this back to our global condition (26) critical value of the fifth which satisfies the tau condi-
and (29), we can find the maximum turbulent luminosity tion. For example, in the spirit of Burrows & Goshy
(1993), we fix MN S , RN S , Tν , vary Ṁ, and find a criti-
Lν τ cal Lν . As is detailed in Murphy & Dolence (2017), this
Lmax
e ≈   . (34)
R r−rg Φρ dr
1 + (γ − 1) tanh h P r
critical curve corresponds to where Ψmin = 0 (which also
corresponds to vs = 0).
Within the context of our assumptions, equation (34) Previously, calculating this critical point only required
is our final algebraic expression which gives the scale local terms in the equations. Since our turbulence model
of turbulence in terms of the background structure and relies on global quantities (Mg , Wb , τ ), we must spec-
the driving neutrino power. The second term in the ify realistic values for these global quantities initially,
denominator is a weak function of shock radius and is then use the density and temperature profiles from the
of order unity. Therefore, the turbulent luminosity is a previous iteration to calculate them for the following
fraction of the neutrino-driving power Lmaxe ≈ Lν τ /(1+ pseudo-solutions. This method is a valid approximation
f (Rs )), and the turbulent dissipation is also a fraction of since the boundaries and constituents of each integral
the driving neutrino power Ek ≈ Lν τ f (Rs )/(1+f (Rs )). varies a negligible amount after each step.
Though we have successfully closed our set of equa-
tions, we did employ several assumptions about the local 2.5. A New Upper Bound on the Reynolds Stress
structure of turbulence. Many of the assumptions made We have discovered that there is an upper bound
in this section have been verified by simulations indi- on the Reynolds stress in the core-collapse problem.
vidually (Murphy et al. 2013) and, since we have been This comes from the fact that the Reynolds stress ap-
careful in being self-consistent, should have equal valid- plies a turbulent ram pressure at the shock, and for
ity when combined. However, there are some assump- large enough shock radii, there is an upper limit to the
tions that require further verification, and in section 2.4, Reynolds stress that allows solutions to the stalled shock
we discuss these details. jump conditions. Of course, the Reynolds stress depends
upon the driving neutrino power, but otherwise, we find
2.4. Solution Method Including Turbulence
that as one increases the shock radius, the Reynolds
We seek steady-state solutions for the background flow stress only goes up slightly; however, the scale of the
including turbulence equations (13-15). This is a global ram pressure at the shock is set by the gravitational po-
boundary value problem, and in this section, we describe tential energy, which decreases as 1/r. Eventually, the
our solution strategy. In essence, the equations for the ratio of the Reynolds stress to the gravitational poten-
turbulence model represent a boundary value problem tial becomes so large that there are no longer solutions
for turbulence embedded within the larger boundary- to the steady-state jump conditions. We now derive the
value problem of the background solution. As is stan- analytic upper bound to the Reynolds stress and de-
dard, to find the steady-state solutions including turbu- scribe our implementation of this limit into our solution
lence, we represent this boundary value problem as a set method.
of coupled first order differential equations and use the To quantify this upper bound, we start with our three
shooting method to find the global solution. jump conditions (7-9) that include turbulence, and make
To ”shoot” for our solution, we first designate the some assumptions so that we can derive an analytic ex-
boundary conditions at the shock for temperature, pres- pression. Our first assumption is that the fluxes of in-
sure, and density. We then integrate inward to the neu- ternal energy and kinetic energy at the shock are small
tron star surface and apply our final boundary condition. relative to the other terms (Meakin & Arnett 2007, 2010;
To find the steady-state stalled solution, the density at Murphy & Meakin 2011b). Our new decomposed bound-
the inner boundary should satisfy τ = 2/3. In the cases ary conditions are thus:
that the density does not match the inner boundary,
then these “solutions” represent pseudo steady-state so- P1 + ρ1 u21 + ρ1 Rrr = P2 + ρ2 u22 (35)
lutions, and the ratio of the inner density to the de- and
sired inner density provides an indication of whether
P1 1 P2 1
the shock would move out or in (see equations (18-21) + e1 + u21 + Rrr ≈ + e2 + u22 (36)
of Murphy & Dolence (2017)). ρ1 2 ρ2 2
Now that we have suitable boundary conditions, we Here we have omitted the 0 subscript, where all non-
can find the cases for which steady-state solutions exist. perturbed terms are implied to be the background.
Since we have several free physical parameters (Lν , Ṁ, Since all of the perturbed components have either can-
MN S , RN S , Tν ), we fix three of them at reasonable val- celed or been defined, there is no need to differentiate
ues, iterate over a fourth parameter, and calculate the with a 0 or 0
 9

Using a γ-law equation of state we can solve for a 2.6. Discussion, Parameters, & Limitations of the
solution of the ratio of densities: Method
γ+1 The approach that we outline in this manuscript pro-
β= (37)
vides a unique way to investigate how turbulence affects
q
γ+ M−2
2 − (1 − M−2
2 )
2 − (γ + 1) 2R rr
u22
the critical condition for explosion, but it will require
Where β is the compression factor, β = ρ1 /ρ2 . For further validation. At the moment, this approach is
physical solutions of β, the term under the radical needs more of a proof of concept; to make it more quanti-
to be positive. This sets an upper limit on the second tative and predictive, there are some parameters and
term which is an upper limit on the Reynolds stress. limitations that must be explored and calibrated with
We now make some approximations to derive a simple, more realistic multi-dimensional simulations. The pri-
analytic limit for the Reynolds stress. If we assume that mary parameters are associated with the size of the con-
the velocity of in-falling matter onto the shock is roughly vective region, L in eq. (22), and the length scale for the
in free fall: turbulent luminosity, h in eq. (32).
1 2 Since the relation between the Reynolds stress and
v ≈ Φ(rs ) (38) turbulent dissipation is modified solely by the length
2 2
scale of convection, it is imperative to treat L prop-
an upper limit on our Reynolds stress becomes
erly. Kolmogorov (1941) argued it to be the size of the
(1 − M−2
2 )
2 largest eddies formed. Since neutrino-driven convection
Rrr ≤ Φ(rs ) , (39) and SASI both exhibit eddies and sloshing on the order
γ+1
of the gain region (Couch & O’Connor 2014; Foglizzo
or in dimensionless form: et al. 2015; Fernández 2015b; Radice et al. 2016), tak-
(γ + 1)Rrr ing the full length of the gain region is not disingen-
R= ≤ 1. (40) uous. Furthermore, simulations show that the inertial
Φ(rs )(1 − M−2 )2
range scaling spans several orders of magnitude, so even
Above this limit, there can be no stalled shock solu- the largest eddies should contribute appropriately to the
tions, and since our method takes this assumption as conversion of kinetic energy to heat, assuming fast cas-
an intermediate step, above this limit, we can not find cade times (Armstrong et al. 1995).
these quasi-steady solutions. Thus, we terminate our Ψ Contrarily, there is little work done in developing an
curve at the radius for which our R parameter crosses analytic turbulence model for core-collapse, thus finding
this threshold. In practice, we had numerical difficulty an appropriate length scale for which the turbulent lu-
when R got close to one. So to avoid that numerical minosity is relevant becomes another parameter of our
difficulty, we set a cap of 0.6 of this value (see fig. 4). problem. For the sake of consistency, only one value of h
This upper limit on the Reynolds stress could have af- has been used throughout this paper. However, varying
fected the critical curve, however it does not. To reiter- values of h yield the same characteristic results (within
ate: the critical neutrino luminosity curve is determined realistic lengths).
by the point of the minimum of the Ψ curve. Theoret- Moreover, since the majority of multi-dimensional ef-
ically, if the imposed upper limit on R were to be at fects are confined to the gain region, we approximate the
a shock radius smaller than our Ψmin , an entirely new effects to be zero below the gain radius and above the
critical curve might have to be defined in order to en- shock. Additionally, simulations have shown that the in-
sure that the above constraint was not being violated. crease in entropy due to turbulence is seen to be strictly
Luckily, in all cases where an actual cap is necessary, within the gain region, further supporting our isolation
this happens to be at a shock radius greater than Ψmin . of the additional heating to the gain region (Murphy
Thus, the cutoff point that we use, ∼60%, is synony- et al. 2013).
mous with the condition of Eq. (39). Though imposing In general, we started with an integral model for tur-
an upper limit on Rrr would ostensibly mitigate its af- bulence, but for the solutions, we require local solu-
fect on the critical curve, we have just shown that all tions and made some significant approximations. For
pseudo-solutions after Ψmin are irrelevant. Hence, we the most part, these approximations seem to be con-
are only constraining the Reynolds stress for pseudo- sistent with multi-dimensional simulations. To validate
solutions which are already non-steady state. That said, these assumptions, the community will need to test these
it is possible that once we start looking at the explod- assumptions with multi-dimensional simulations.
ability of a set of initial conditions, our upper limit may
start to interfere with predictions. This upper limit will
be treated accordingly with more rigor in future publi-
cations. 3. RESULTS & DISCUSSION
 10

2.5 Without Turbulence

2.0 gain radius With Turbulence
shallower gradient from turbulence
1.5 1010

(g/cc)
1.0
Terms for dlnρ
dlnr

0.5
0.0 109
0.5 net ν heating
1.0 Turbulent Dissipation
Turbulent Luminosity 70 80 90 100 110
1.5 Ram Pressure Term Radius (km)
2.0 100 150 200 250 2.75 Without Turbulence
Radius (km) 2.50 With Turbulence
Figure 1. The effects of neutrinos and turbulence on the 2.25
density and temperature profiles. Specifically, we show the
2.00

MeV
dimensionless terms that appear in the equation for dd ln ρ
,
A1. We omit the pressure and gravity terms which combine
ln r
1.75
to give a power-law slope of about −3. Instead we show the
net neutrino heating and cooling term (solid-green line) and
1.50
the effect of each turbulent term. The turbulent dissipation 1.25
(solid, blue line) and the turbulent luminosity (dashed, blue
line) terms originate from the energy equation, and the ram 1.00
pressure term (dotted, red line) comes from the momentum 70 80 90 100 110
equation. In our model, we assume that turbulence is driven
only in the gain region. In general, the turbulent terms asso-
Radius (km)
ciated with the energy equation are larger than ram pressure.
More specifically, turbulent dissipation generally affects the Figure 2. The density and temperature profiles for stalled
profile more than the ram pressure. shock solutions with and without the convection. Generally,
convection causes shallower gradients and higher tempera-
tures. While technically all of the terms contribute to this
effect, the most dominant term is turbulent dissipation which
Our primary objective is to understand how turbu- provides extra heating.
lence affects the conditions for explosion. To fully un-
derstand this influence, we also need to understand how see equation (A1). In general, the missing terms give a
turbulence affects the background structure, so we first slope that is about -3. In the gain region, both neutri-
show how the turbulent terms affects the density and nos and the convective terms make the slope shallower.
temperature profiles. We then show that turbulence In the cooling region, neutrino cooling makes the slope
raises the Ψ parameter, implying an easing of the ex- steeper. Turbulent dissipation and luminosity terms ul-
plosion condition. We then consider how this affects timately originate from the energy equation (3). The
the critical hypersurface. To connect to previous in- turbulent ram pressure term comes from the momen-
vestigations, we focus on the neutrino-luminosity and tum equation. While turbulent ram pressure does mod-
accretion-rate slice of this critical hypersurface. We find ify the density structure, the two turbulent terms from
that this reduction in the critical curve is ∼30%, in con- the energy equation, turbulent dissipation and turbu-
cordance with multi-dimensional simulations. To inves- lent luminosity, have a considerably larger effect on the
tigate how turbulence reduces the condition for explo- structure.
sion, we calculate the critical condition with each term Figure 2 shows how turbulence affects the density and
included and omitted. Lastly, we provide evidence that temperature profiles of the steady-state stalled shock
our upper limit on the Reynolds stress does not affect solutions. The net effect of turbulence is to make the
the actual reduction of the critical curve. profiles much shallower. In part, turbulent ram pres-
In Figure 1, we illustrate how turbulence affects the sure explains some of the difference, but for the most
density profile; in particular, we show the neutrino and part, turbulence provides extra heating through dissi-
convective terms in the derivative for the density. Since pation in the convective region. One consequence is
the density profile most closely matches a power-law, that the temperature (and entropy) are higher with tur-
we plot terms of d ln ρ/d ln r to compare how neutrinos bulence. This is consistent with the entropy profiles
and turbulence affect the slope in the log. To reduce the of multi-dimensional simulations (Murphy & Meakin
clutter, we do not plot all of the terms; for the full ODE, 2011a; Murphy et al. 2013). In agreement with simu-
 11

5 Without Convection, Lν = 2
With Convection, Lν = 2
4 Lν = 2. 1
Lν = 2. 2
3 Lν = 2. 3
Lν = 2. 4
Lν = 2. 5
2 Indicates the largest shock radius

Ψ
which yields a physical solution

1
0
1 3 4 5 6 7 8
xs
Figure 3. Ψ parameter as a function of shock radius, xs = 0.6
Rs /RNS . Ψ is roughly the over pressure compared to hy-
drostatic equilibrium normalized by the pre-shock ram √ pres-
sure. The shock velocity is related to Ψ by vs ≈ 1 − 1 + Ψ.
0.5
Therefore, the sign of Ψ determines whether the shock re-
0.4
(1 − M −2 ) 2
cedes, expands, or is stalled. There is always a minimum for

γ+1
Ψ, Ψmin . If Ψmin < 0, then there is always a stalled solution.
On the other hand, when Ψmin > 0, then only vs > 0 solu-
tions exist. Murphy & Dolence (2017) propose that Ψmin = 0
R = Rφ 0.3
is the explosion condition. Adding turbulence has the effect Lν = 2
of raising Ψmin , making it easier to reach the explosion con- 0.2 Lν = 2. 1
dition. Adding turbulence has a similar effect as increasing Lν = 2. 2
the neutrino luminosity by 30% (red line).
0.1 Lν = 2. 3
Lν = 2. 4
Lν = 2. 5
lations (Murphy et al. 2013; Abdikamalov et al. 2016), 0.0 3 4 5 6 7 8
Figure 2 also shows that the solution including turbu- xs
lence has a larger shock radius. The shock radius is a
monotonic function of Lν , and since ρRrr and ρ effec- Figure 4. Ψ and corresponding dimensionless Reynolds
tively add energy in the same fashion as the luminosity, stress, R, as a function of shock radius (xs = r/rs ). There is
an upper limit for the dimensionless Reynolds stress which
we intuitively retrieve a larger shock radius. Note that we derive in section 2.5 (see equation (40)). Here, we show
this larger shock radius is not just a consequence of tur- the behavior of the normalized Reynolds stress vs. xs and
bulent ram pressure. Instead, the post shock profile is how it affects the explodability parameter Ψ. In the bot-
tom panel, each R increases until it reaches an unphysical
shallower pushing out the shock. point and terminates at the red dot. The same termination
Now that we understand how turbulence affects the points can be seen above in the top panel, where each dot
density and temperature profiles, we now present how corresponds to its respective unphysical shock radius for a
turbulence affects the critical curve in three figures. given Lν . If the cap had occurred to the left of Ψmin then
this upper bound on R would have affected the critical curve.
One, Figure 3 shows how turbulence raises the dimen- However, the upper limit occurs to the right of Ψmin , there-
sionless overpressure parameter, Ψ, in Murphy & Do- fore it does not affect the critical curve. Thus, the critical
lence (2017). Two, Figure 5, shows how turbulence re- point of explosion is still dominated by non-ram pressure
terms.
duces the five-dimensional critical hypersurface for ex-
plosion Finally, Figure 7 shows that the dominant turbu-
lent term in reducing the critical condition is turbulent shock expands, when Ψ is negative, the shock stalls,
dissipation. and when Ψ = 0, the shock has zero velocity. Note that
In Figure 3 we plot Ψ to show that the increase in the there is always a minimum Ψ, and if this Ψmin is greater
dimensionless parameter due to turbulence is roughly than zero, then there are no stalled shock solutions, only
equivalent to increasing the neutrino luminosity by 30%. steady expanding shocks. For the case where the mini-
To clarify, the Ψ parameter is a measure of the over- mum of Ψ is exactly zero, this set of solutions defines our
pressure compared to the hydrostatic equilibrium. This critical explosion condition for all parameters. Clearly
integral condition is normalized by the pre-shock ram turbulence raises the minimum Ψ, and therefore would
pressure. This dimensionless parameter maps directly affect the critical curve.
to the shock velocity in that when Ψ is positive, the In figure 4, we show how the Reynolds stress upper
12

Figure 5. How turbulence affects the Ψmin = 0 explosion condition. Ψmin depends upon five parameters of the core-collapse
problem: the neutrino luminosity, Lν , mass accretion rate, Ṁ, neutron star radius or more specifically the neutrino-sphere
radius, Rν , the neutrino temperature, Tν , and the neutron star mass, MNS . Therefore, the explosion condition Ψmin > 0
represents a hypersurface in this five-dimensional parameter space. By fixing 3 of the 5 parameters, one may construct “critical
curves” with the other 2 parameters. The critical Lν − Ṁ (lower left panel) is one such example. In each panel, the solid line
shows the critical condition Ψmin = 0, and for all but the top panel explosions occur in the upper portion of the parameter
space. The dashed line shows the reduction of the critical condition due to neutrino-driven convection for each critical curve.

limit affects the explodability parameter and the criti- the critical condition for explosion is not a critical curve,
cal curve. We suspected that, at high enough neutrino but a critical hypersurface in a five-dimensional space
luminosities, the additional ram pressure at the shock that is defined by one dimensionless condition: Ψmin =
would prevent finding solutions to the boundary condi- 0. The neutrino-luminosity-accretion-rate curve is one
tions. Figure 4 demonstrates that we consistently en- slice of this critical hypersurface. In Fig. 5, we show six
counter this cap, but only for pseudo-solutions which slices of this hypersurface. Note that in all panels except
have already found a critical Lν . We present several sets the top-left panel (MNS vs. Ṁ), the region of explosion
of solutions at different neutrino luminosities to empha- is above the curve. For MNS vs. Ṁ, it is below the
size that our Lν − Ṁ critical curve is in fact not affected curve; a lower mass neutron star has a lower potential
by the upper limit on R, even at unrealistic luminosi- to overcome to explode. Because turbulence raises the
ties. The sole determination of the critical curve is on Ψ curves, it also reduces the critical hypersurface for
Ψmin = Ψ(rscrit ), and since this threshold is only reached explosion.
when rs > rscrit , there is no effect on the critical curve. In figure 6, we make a case for the validity of using a
Figure 5 shows how turbulence modifies the critical convection-based turbulence model. The χ parameter,
hypersurface. Murphy & Dolence (2017) points out that first introduced by Foglizzo et al. (2006), is a measure of
 13

in simulations to validate wether convection is indeed

5 All Turbulent Effects dominant near the critical condition.
Original Critical Curve
In figures 3 and 5, we considered the overall effect of
4 =3
turbulence on the critical condition, but this does not
L [1052 erg/s]

illuminate how turbulence reduces the condition for ex-

3 plosion. In figure 7, we focus exclusively on the Lν − Ṁ
critical curve and explore the effect of each term in this
2
Convection slice of the critical condition. We find that turbulent dis-
sipation within the gain region acts as an even greater
1
SASI driving force for explosion than the turbulent ram pres-
sure. That is not to say that the Reynolds stress is
0
0.2 0.4 0.6 0.8 negligible; both terms are indeed needed to obtain the
[M /s] critical curve reduction predicted by multi-dimensional
simulations. Though this result relies upon some as-
sumptions in the turbulence model, we suspect that the
Figure 6. Comparing the relative thresholds of the criti- qualitative outcome will persist: turbulent dissipation
cal curves and the χ = 3 line. Above this line, all stalled can not be dismissed.
solutions have χ > 3 (convection dominated), and below
this line, χ < 3 (SASI dominated). The fact that convec-
tion dominates near the critical curve validates our use of 3.1. Buoyancy Driven Heating
a convection-based turbulence model to explore how turbu- We argue that turbulent dissipation adds another heat
lence affects the critical condition for explosions.
source which aids explosion. Since neutrinos are the ul-
timate source of energy, it may seem that we are asking
the linear stability of the convective region in the pres- neutrinos to do twice the work. However, this is not the
ence of advection. For χ < 3, advection stablizes the case. Using a simple convective model, we propose that
flow and convection does not mainfest. The assumption in one dimension, some of the neutrino energy goes into
is that since convection is suppressed, the SASI dom- heating the material, and some of it goes into creating
inates the turublence. The χ parameter is defined as a higher potential profile. In multiple dimensions, this
higher potential profile is unstable, goes to a lower po-
Z rs
ωbuoy tential profile, and the excess energy goes into kinetic
χ≡ dr (41)
rg u energy, which in turn dissipates as heat via turbulent
dissipation.
where ωbuoy is the Brunt-Väisälä frequency, defined as Rayleigh-Bénard convection is a simple convective
model which can clearly demonstrate this conversion
"     #
2 1 ∂p dS 1 ∂p dY e g
ωbuoy ≡ + from potential to kinetic to dissipated internal energy.
p ∂S ρ,Ye dr p ∂Ye ρ,S dr Γ1
Figure 8 illustrates the fundamental physics of convec-
(42)
tion. First, neutrinos provide a source of heating and
and 
∂ ln P
 drives a convective instability. In this cartoon model, we
Γ1 ≡ . (43) consider two parcels; the lower one receives more neu-
∂ ln ρ S,Ye
trino heating, has a higher entropy, and lower density
In Figure 6, above the χ = 3 dashed line, all stalled solu- compared to its surroundings. The parcel at a higher
tions have χ > 3 (convection dominated) and below this height has a lower entropy and higher density compared
dashed line, all stalled solutions have χ < 3 (SASI domi- to its surroundings. If one switches the positions of these
nated). To calculate this line, we use a similar approach two parcels, then one finds that the gravitational po-
as deriving the critical luminosity curves: we input all tential is lower. Therefore, neutrino heating causes a
of our parameters, and numerically solve for the lumi- higher potential structure that is unstable to convective
nosity which gives a value of χ = 3. According to these overturn. The difference in potentials between the two
models, convection dominates for all scenarios near ex- states gets converted into kinetic energy of the parcels.
plosion. Recent simulations seem to support this conclu- Kolmogorov’s hypothesis suggests that the dissipation
sion (Abdikamalov et al. 2015; Lentz et al. 2015; Couch of this kinetic energy is R3/2 /L and happens on the or-
& O’Connor 2014; Burrows et al. 2012; Iwakami et al. der of one turnover timescale. Burrows et al. (1995)
2014; Ott et al. 2013; Roberts et al. 2016). The results also considered this idea in which two layers of varying
of Figure 6 validate our exploration of how convection- densities are swapped, inducing a buoyant work being
based turbulence affects the critical condition. However, done on the system. This energy is then converted into
we do encourage future explorations of the SASI and χ kinetic energy in the form of eddies, and in turn dissi-
14

5
Ψ min > 0 & Only
4 vs > 0 solutions, Explosions
Lν [10 52 erg/s]
3
2
Ψ min < 0
All Turbulent Effects
1 Stalled Solutions Original Critical Curve
Only TD & TL
Only Reynolds Stress
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Ṁ [M ¯ /s]

Figure 7. Diagnosing how turbulence affects a critical curve for explosion. Here we show how the various terms affect one
particular slice of the Ψmin = 0 condition, the neutrino-luminosity vs. mass-accretion-rate critical curve. The thick red line is
the original critical curve (Burrows & Goshy 1993), and the thick blue line shows the turbulence induced reduction in the critical
curve. The red-shaded region is where Ψmin > 0 and thus where no steady state solutions exist. The blue-shaded region is the
region of stalled shock solutions when including convection. To assess the effects of each of the turbulence terms, we reproduce
the critical curve by isolating the turbulent terms in the energy and momentum equations. In the energy equation, the terms
are the turbulent dissipation and turbulent luminosity (TD and TL), and in the momentum equation, the only turbulent term
is the Reynolds stress (or turbulent ram pressure). The dotted line corresponds to only including Reynolds stress, and the
dashed line corresponds to only including the effects of TD and TL. From these results we draw two main conclusions. One,
the necessary neutrino energy required for a supernova explosion is less when considering multi-dimensional effects. Two, most
of the reduction in the critical condition comes from the energy equation terms, in particular the turbulent dissipation.

pated into heat. Therefore, not only do neutrinos heat buoyantly rises, the potential is then
the gain region, but they also create a higher potential
Φ2 = gV (ρ2 h1 + ρ1 h2 ) (46)
system. This higher potential gets converted to kinetic
energy and consequently dissipated as internal energy. Thus, the change in potential is
To illustrate this more clearly, consider the energy
equation (3). It is more illuminating if we rewrite the ∆Φ = gV (ρ2 − ρ1 )(h1 − h2 ) (47)
equation considering a constant mass accretion rate: Since h1 > h2 and ρ1 > ρ2 , ∆Φ < 0. Conservation of
energy suggests that K.E.≈-∆Φ, and via turbulent dis-
uj uj
 
Ṁ∂i h + − Φ = ρq (44) sipation, this energy from buoyant driving is converted
2 into heat. In summary, neutrinos heat the gain region
Neutrinos heat the convective region of the star, chang- and setup a higher potential profile. Turbulence allows a
ing the enthalpy and gravitational potential. This gives lower potential state and the kinetic energy is converted
rise to an entropy and density gradient such that S2 > into thermal energy via turbulent dissipation, aiding ex-
S1 and ρ2 < ρ1 (thus satisfying the Schwarzschild con- plosion.
dition for convection). We then treat the potential en-
4. CONCLUSION
ergy of two parcels in a similar manner to Burrows et al.
(1995). Before the exchange of parcels, the gravitational A major result of core-collapse theory is that one-
potential energy is dimensional simulations fizzle for all but the least mas-
sive stars, while multi-dimensional simulations explode
Φ1 = gV (ρ1 h1 + ρ2 h2 ) (45) for the most part. Even though there are still some
differences between the simulations, there is a general
But after the top parcel sinks and the bottom parcel consensus that turbulence makes the difference between
 15

1
2

1 2

2 1

1
a) b) c)

Figure 8. Neutrinos heat the gain region and setup a higher potential state, which turbulence taps and dissipates as heat. Panel
a): when a parcel of matter advects through the gain region, neutrinos heat it, which sets up a buoyantly unstable situation.
Panel b): Parcel 1 wants to buoyantly sink and parcel 2 wants to buoyantly rise. The final state has a lower potential energy
than the final state. Panel c): This change in potential energy is converted to turbulent kinetic energy which dissipates via
turbulent dissipation. Therefore, in 1D, part of the energy of neutrinos heats the gain region and part of the neutrino energy
goes into setting up the higher potential profile. Multi-dimensional turbulence taps into this higher potential energy by allowing
for a lower potential state and turbulent kinetic energy. Then that turbulent kinetic energy is dissipated through viscosity.

a fizzled outcome and a successful explosion. To ex- terms, both in the background solution and in the
plain how turbulence aids explosion, we develop a tur- boundary conditions. The three main turbulent terms
bulence model for neutrino-driven convection (Murphy are turbulent dissipation, Reynolds stress, and turbulent
et al. 2013) and investigate how turbulence reduces the luminosity. Overall, we find that all three play an impor-
critical condition for explosion (Burrows & Goshy 1993; tant role in modifying the background structure and the
Murphy & Dolence 2017). Our turbulence model re- critical condition for explosion. However, it is the terms
duces the critical condition for explosion by ∼30%, in in the energy equation, the combined turbulent lumi-
general agreement with simulations. By modeling each nosity and turbulent dissipation, which give the largest
turbulent term, we are able to investigate the effect of reduction in the critical curve. Of these two, the tur-
each turbulent term on the critical condition. We find bulent dissipation provides the largest effect in reducing
that although ram pressure plays a role in reducing the the critical condition.
critical curve, it is not the dominant term. The dom- The ultimate source of power for convection and tur-
inant term in reducing the critical curve is turbulent bulent dissipation is the neutrino driving power. This
dissipation. Furthermore, we are not the first to sug- may seem as if we are double counting the power sup-
gest that turbulent dissipation is important in aiding plied by neutrinos. However, we are not. Instead, we
neutrino-driven explosions. Thompson et al. (2005) sug- propose that neutrinos heat the gain region and setup
gested that MRI driven turbulence for very rapidly ro- a higher potential, convectively unstable structure. In
tating proto-neutron stars could add significant heating multi-dimensional simulations, convection converts this
and aid explosion. higher potential structure to a lower potential struc-
In the turbulence model, we include all turbulent ture. The change in potential energy is converted to ki-
16

netic turbulent energy, which in turn dissipates as heat. a higher potential compared to their multi-dimensional
We suggest that CCSN modelers check the structures counterparts? Third, do the local details of the turbu-
of their one-dimensional and multi-dimensional simula- lent model matter? We suspect that the local details do
tions to test this supposition. not change the qualitative result. However, one should
The turbulence model is a global, integral model, and compare our local assumptions with multi-dimensional
provides little constraint on the local structure of tur- simulations, and assess how (if at all) the quantitative
bulence, but a local model is necessary in solving the results vary from this work.
background equations and deriving a critical condition. In summary, combining a turbulence model and crit-
Therefore, we made some fairly straightforward assump- ical condition analyses helps to illuminate how turbu-
tions to translate the global model to a local model. For lence aids explosions. Specifically, by modeling each
example, we assumed that the specific turbulent dissipa- turbulent term we are able to assess the effect of each
tion rate is constant throughout the gain region. We also term in the conditions for explosion. Contrary to prior
had to assume a specific spatial profile for the turbulent suppositions, we find that turbulent ram pressure is not
luminosity. These assumptions probably do not affect the dominant effect. Rather, each of the three terms
our qualitative results. However, our results have large are quite large in their effect with turbulent dissipation
implications regarding how turbulence aids explosion. being the largest. Presently, these conclusions are qual-
In particular, we propose that turbulent dissipation is a itative. To be predictive, the community will need to
key contributor to reviving a stalled shock to a success- verify these conclusions with multi-dimensional simula-
ful supernova. Therefore, these assumptions shoud be tions. Eventually, we may be able to use these turbu-
verified with multi-dimensional simulations and treated lence models to make one-dimensional simulations ex-
more rigorously in future investigations. plode under similar conditions as multi-dimensional sim-
To verify the predictions of this manuscript, we iden- ulations, thereby enabling rapid and systematic explo-
tify at least three open questions that multi-dimensional ration of the explosion of massive stars.
simulations should address. One, does turbulent dis-
sipation actually lead to significant heating? In mul- ACKNOWLEDGMENTS
tiple dimensions, the entropy profile should be a re- We would like to thank Joshua C. Dolence for illumi-
sult of neutrino heating and cooling, turbulent entropy nating discussions on the effect of turbulence on the crit-
flux, and turbulent dissipation. One can easily calculate ical condition. We are also thankful to David Radice for
the turbulent entropy flux, and the heating and cool- challenging us to explain the source of energy for turbu-
ing by neutrinos. What ever is left should be equal lent dissipation. This material is based upon work sup-
to the expected entropy generated by turbulent dissi- ported by the National Science Foundation under Grant
pation. Second, do the one-dimensional profiles have No. 1313036.

APPENDIX
A. FULL ODES
Here, we present the full set of ordinary differential equations that we solve to find pseudo steady-state solutions.
The equation for density is
∂lny

2v 2 2
Φ −1+RS+ yl ∂lnT
∂lnl 1+H−C+T D−T L+ 2v
Φ
dlnρ
dlnr = ∂lnT
∂lny
, (A1)
2 yv 2
y− vΦ +y ∂lny
∂lnρ +
∂lnT
∂lnl Φl
∂lnl
− ∂lnρ
∂lnT

and in terms of the density equation, the temperature ODE is

2

 
∂ ln T 1+H−C+T D−T L+ 2v
Φ
∂lnρ
v2 ∂lnl
∂ ln r = ∂lnl
l ∂lnT
+ ∂lnr
∂lnl Φl − ∂lnρ . (A2)
∂lnT

In this form, these equations seem some what unwieldy, but they would be even more so if we had not made the
following shorthand for the important physics. To further help illustrate the important scales in the problem, we
present each important physics in terms of a dimensionless variable. The Reynolds stress (or ram pressure) appears
in the above equations as
2x2 Rrr
RS = . (A3)
(xs − xg )Φ1
In these expressions, the subscript 1 indicates the base of the solution. In our particular case, that is the neutrino-sphere
 17

radius. Neutrino heating and cooling are

Lν κρr1 x2
H= , (A4)
ṀΦ1
and
T 6


C0 T0 4πr13 x4 ρ
C= . (A5)
ṀΦ1
The two turbulent terms from the energy equation are the turbulent dissipation and turbulent luminosity:
Lmax
e r13 ρwb 4π
TD= (A6)
Mg ṀΦ1
and
Lmax
e r1 x2
TL= (r1 (x−xg ))

. (A7)
cosh2 h hṀΦ1
Two dimensionless measures of the pressure and enthalpy are
P
y= (A8)
ρΦ
and
P
e+ ρ
l= (A9)
Φ

∂ ln P
P,ρ = (A10)
∂ ln ρ T

The normalized radius is

r
x= . (A11)
rN S
A dimensionless measure of the bouyant driving is
Zxs r1 (x−xg )

tanh h
wb = dx . (A12)
3yx
xg

To observe and distinguish the effects of each turbulent term, we add the capability to turn each term on or off in the
equations. In doing so, we investigate the effect of each term on the solutions and critical curves.

REFERENCES
Abdikamalov, E., Zhaksylykov, A., Radice, D., & Berdibek, S. Burrows, A., & Goshy, J. 1993, ApJL, 416, L75
2016, ArXiv e-prints, arXiv:1605.09015 Burrows, A., Hayes, J., & Fryxell, B. A. 1995, ApJ, 450, 830
Abdikamalov, E., Ott, C. D., Radice, D., et al. 2015, ApJ, 808, Burrows, A., Vartanyan, D., Dolence, J. C., Skinner, M. A., &
70 Radice, D. 2016, ArXiv e-prints, arXiv:1611.05859
Armstrong, J. W., Rickett, B. J., & Spangler, S. R. 1995, ApJ, Canuto, V. M. 1993, ApJ, 416, 331
443, 209 Colgate, S. A., & Herant, M. E. 2004, in Cosmic explosions in
Bethe, H. A. 1990, Reviews of Modern Physics, 62, 801 three dimensions, ed. P. Höflich, P. Kumar, & J. C. Wheeler,
Bethe, H. A., & Wilson, J. R. 1985, ApJ, 295, 14 199
Blondin, J. M., Mezzacappa, A., & DeMarino, C. 2003, ApJ, Couch, S. M. 2013, ApJ, 775, 35
584, 971 Couch, S. M., & O’Connor, E. P. 2014, ApJ, 785, 123
Bruenn, S. W., Lentz, E. J., Hix, W. R., et al. 2016, ApJ, 818, Couch, S. M., & Ott, C. D. 2015, ApJ, 799, 5
123 Dolence, J. C., Burrows, A., Murphy, J. W., & Nordhaus, J.
Buras, R., Janka, H.-T., Rampp, M., & Kifonidis, K. 2006a, 2013, ApJ, 765, 110
A&A, 457, 281 Dolence, J. C., Burrows, A., & Zhang, W. 2015, ApJ, 800, 10
—. 2006b, A&A, 457, 281 Ertl, T., Janka, H.-T., Woosley, S. E., Sukhbold, T., & Ugliano,
Buras, R., Rampp, M., Janka, H.-T., & Kifonidis, K. 2003, M. 2016, ApJ, 818, 124
Physical Review Letters, 90, 241101 Fernández, R. 2015a, MNRAS, 452, 2071
Burrows, A., Dessart, L., Ott, C. D., & Livne, E. 2007, PhR, —. 2015b, MNRAS, 452, 2071
442, 23 Fernández, R., Müller, B., Foglizzo, T., & Janka, H.-T. 2014,
Burrows, A., Dolence, J. C., & Murphy, J. W. 2012, ApJ, 759, 5 MNRAS, 440, 2763
18

Foglizzo, T., Scheck, L., & Janka, H.-T. 2006, ApJ, 652, 1436 Mazurek, T. J., Cooperstein, J., & Kahana, S. 1982, in NATO
Foglizzo, T., & Tagger, M. 2000, A&A, 363, 174 Advanced Science Institutes (ASI) Series C, Vol. 90, NATO
Advanced Science Institutes (ASI) Series C, ed. M. J. Rees &
Foglizzo, T., Kazeroni, R., Guilet, J., et al. 2015, PASA, 32, e009
R. J. Stoneham, 71–77
Guilet, J., Sato, J., & Foglizzo, T. 2010, ApJ, 713, 1350 Meakin, C. A., & Arnett, D. 2007, ApJ, 667, 448
Handy, T., Plewa, T., & Odrzywolek, A. 2014, ApJ, 783, 125 Meakin, C. A., & Arnett, W. D. 2010, Ap&SS, 328, 221
Hanke, F., Marek, A., Müller, B., & Janka, H.-T. 2012, ApJ, Melson, T., Janka, H.-T., & Marek, A. 2015, ApJL, 801, L24
755, 138 Müller, B. 2016, ArXiv e-prints, arXiv:1608.03274
—. 2017, ArXiv e-prints, arXiv:1702.06940
Hanke, F., Müller, B., Wongwathanarat, A., Marek, A., & Janka,
Müller, B., Heger, A., Liptai, D., & Cameron, J. B. 2016,
H.-T. 2013, ApJ, 770, 66 MNRAS, 460, 742
Herant, M., Benz, W., Hix, W. R., Fryer, C. L., & Colgate, S. A. Müller, B., Melson, T., Heger, A., & Janka, H.-T. 2017, ArXiv
1994, ApJ, 435, 339 e-prints, arXiv:1705.00620
Hillebrandt, W., & Mueller, E. 1981, A&A, 103, 147 Murphy, J. W., & Burrows, A. 2008, ApJ, 688, 1159
Murphy, J. W., & Dolence, J. C. 2017, ApJ, 834, 183
Horiuchi, S., Beacom, J. F., Kochanek, C. S., et al. 2011, ApJ,
Murphy, J. W., Dolence, J. C., & Burrows, A. 2013, ApJ, 771, 52
738, 154 Murphy, J. W., & Meakin, C. 2011a, ApJ, 742, 74
Iwakami, W., Nagakura, H., & Yamada, S. 2014, ApJ, 793, 5 —. 2011b, ApJ, 742, 74
Janka, H.-T. 2001, A&A, 368, 527 O’Connor, E., & Ott, C. D. 2011, ApJ, 730, 70
—. 2017, ArXiv e-prints, arXiv:1702.08825 Ott, C. D., Abdikamalov, E., Mösta, P., et al. 2013, ApJ, 768,
115
Janka, H.-T., & Keil, W. 1998, in Supernovae and cosmology, ed.
Pejcha, O., & Thompson, T. A. 2012, ApJ, 746, 106
L. Labhardt, B. Binggeli, & R. Buser, 7 Radice, D., Burrows, A., Vartanyan, D., Skinner, M. A., &
Janka, H.-T., Melson, T., & Summa, A. 2016, ArXiv e-prints, Dolence, J. C. 2017, ArXiv e-prints, arXiv:1702.03927
arXiv:1602.05576 Radice, D., Ott, C. D., Abdikamalov, E., et al. 2016, ApJ, 820,
Janka, H.-T., & Müller, E. 1995, ApJL, 448, L109 76
Rampp, M., & Janka, H.-T. 2002, A&A, 396, 361
—. 1996, A&A, 306, 167
Roberts, L. F., Ott, C. D., Haas, R., et al. 2016, ArXiv e-prints,
Kitaura, F. S., Janka, H.-T., & Hillebrandt, W. 2006, A&A, 450, arXiv:1604.07848
345 Sato, J., Foglizzo, T., & Fromang, S. 2009a, ApJ, 694, 833
Kolmogorov, A. 1941, Akademiia Nauk SSSR Doklady, 30, 301 Sato, J., Foglizzo, T., & Fromang, S. 2009b, in SF2A-2009:
Lentz, E. J., Bruenn, S. W., Hix, W. R., et al. 2015, ArXiv Proceedings of the Annual meeting of the French Society of
e-prints, arXiv:1505.05110 Astronomy and Astrophysics, ed. M. Heydari-Malayeri,
C. Reyl’E, & R. Samadi, 175
Li, W., Leaman, J., Chornock, R., et al. 2011, MNRAS, 412, 1441
Tamborra, I., Hüdepohl, L., Raffelt, G. G., & Janka, H.-T. 2017,
Liebendörfer, M., Mezzacappa, A., & Thielemann, F.-K. 2001a, ApJ, 839, 132
PhRvD, 63, 104003 Thompson, C. 2000, ApJ, 534, 915
Liebendörfer, M., Mezzacappa, A., Thielemann, F.-K., et al. Thompson, T. A., Burrows, A., & Pinto, P. A. 2003, ApJ, 592,
2001b, PhRvD, 63, 103004 434
Thompson, T. A., Quataert, E., & Burrows, A. 2005, ApJ, 620,
Liebendörfer, M., Rampp, M., Janka, H.-T., & Mezzacappa, A.
861
2005, ApJ, 620, 840 Yamasaki, T., & Yamada, S. 2005, ApJ, 623, 1000
Marek, A., & Janka, H.-T. 2009, ApJ, 694, 664 —. 2006, ApJ, 650, 291
Mazurek, T. J. 1982, ApJL, 259, L13