Draft version September 1, 2023

Typeset using LATEX twocolumn style in AASTeX631

The nuclear reaction network WinNet

M. Reichert ,1 C. Winteler,2O. Korobkin ,3 A. Arcones ,4, 5 J. Bliss,4 M. Eichler ,4 U. Frischknecht,2 C. Fröhlich ,6
R. Hirschi ,7, 8 M. Jacobi ,4 J. Kuske ,4 G. Martı́nez-Pinedo ,5, 4 D. Martin ,4 D. Mocelj,2 T. Rauscher ,2, 9 and
F.-K. Thielemann 2, 5
1 Departament d’Astonomia i Astrofı́sica, Universitat de València, Edifici d’Investigatció Jeroni Munyoz, C/Dr. Moliner, 50, E-46100 Burjassot (València), Spain
2 Department of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland
3 Center for Theoretical Astrophysics, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
arXiv:2305.07048v2 [astro-ph.IM] 31 Aug 2023

4 Institut für Kernphysik (Theoriezentrum), Technische Universität Darmstadt, Schlossgartenstr. 2, D-64289 Darmstadt, Germany
5 GSI Helmholtzzentrum für Schwerionenforschung GmbH, Planckstr. 1, D-64291 Darmstadt, Germany
6 Department of Physics, North Carolina State University, Raleigh, NC 27695, USA
7 Astrophysics Group, Lennard-Jones Laboratories, Keele University, Keele ST5 5BG, UK
8 Institute for the Physics and Mathematics of the Universe (WPI), University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa 277-8583, Japan
9 Centre for Astrophysics Research, University of Hertfordshire, Hatfield AL10 9AB, United Kingdom

ABSTRACT
We present the state-of-the-art single-zone nuclear reaction network WinNet that is capable of calculating
the nucleosynthetic yields of a large variety of astrophysical environments and conditions. This ranges from
the calculation of the primordial nucleosynthesis, where only a few nuclei are considered, to the ejecta of
neutron star mergers with several thousands of involved nuclei. Here we describe the underlying physics and
implementation details of the reaction network. We additionally present the numerical implementation of two
different integration methods, the implicit Euler method and Gears method along with their advantages and
disadvantages. We furthermore describe basic example cases of thermodynamic conditions that we provide
together with the network and demonstrate the reliability of the code by using simple test cases. With this
publication, WinNet is publicly available and open source at GitHub and Zenodo.

Keywords: methods: numerical — nuclear reactions, nucleosynthesis, abundances

1. INTRODUCTION Woosley 2010; Maeder & Meynet 2012; Frischknecht et al.

Nuclear reaction networks are crucial to investigate the syn- 2016; Thielemann et al. 2018a; Limongi & Chieffi 2018; Ar-
thesis of elements and their isotopes in astrophysical events. nett et al. 2019; Kaiser et al. 2020; Eggenberger et al. 2021),
While the events can vastly differ in their conditions, the pro- and more violent explosive events such as classical novae
cedure to derive their ejecta composition is always similar. (e.g., Arnould et al. 1980; Wiescher et al. 1986; José et al.
The foundation of the understanding of the origin of elements 2004; Jose 2016; Vasini et al. 2022), X-ray bursts (e.g., Wi-
has been outlined already in Alpher et al. (1948), the so-called escher et al. 1986; Rembges et al. 1997; Schatz et al. 1998;
𝛼𝛽𝛾-Paper. Cyburt et al. 2010; Jose 2016; Meisel et al. 2020), type Ia
The field of nucleosynthetic calculations encompasses the supernovae (e.g., Arnett 1969; Arnett et al. 1971; Iben & Tu-
production of the light elements during the Big Bang (e.g., tukov 1984; Nomoto et al. 1984; Woosley et al. 1986; Mueller
Peebles 1966; Wagoner et al. 1967; Yang et al. 1984; Boes- & Arnett 1986; Thielemann et al. 1986; Khokhlov et al. 1993;
gaard & Steigman 1985; Kawano et al. 1988; Olive et al. 1990; Höflich et al. 1998; Röpke et al. 2012; Hillebrandt et al. 2013;
Walker et al. 1991; Smith et al. 1993; Cyburt et al. 2016; Coc Pakmor et al. 2013; Dan et al. 2015; Maeda & Terada 2016;
& Vangioni 2017; Pitrou et al. 2018, 2021; Fields & Olive Garcı́a-Senz et al. 2016; Jiang et al. 2017; Röpke & Sim 2018;
2022), the element production during the lifetime of stars Thielemann et al. 2018b; Shen et al. 2018; Leung & Nomoto
(see, e.g., Kippenhahn et al. 2013; Karakas & Lattanzio 2014; 2018; Gronow et al. 2021; Lach et al. 2022), core-collapse
Karakas & Lugaro 2016; Bisterzo et al. 2017; Kobayashi et al. supernovae (e.g., Kotake et al. 2012; Burrows 2013; Janka
2020; Busso et al. 2021; Doherty et al. 2017; Gil-Pons et al. et al. 2016; Müller 2016; Radice et al. 2018; Müller 2020;
2018; Leung & Nomoto 2018; Leung et al. 2020; Arnett Vartanyan et al. 2022) with a focus on nucleosynthesis (e.g.,
1977; Woosley & Weaver 1995; Heger et al. 2003; Heger & Woosley & Weaver 1995; Thielemann et al. 1996; Woosley
& Heger 2006; Heger & Woosley 2010; Perego et al. 2015;
2

Sukhbold et al. 2016; Wanajo et al. 2018; Curtis et al. 2019; For astrophysical scenarios with a much larger diffusion
Witt et al. 2021; Ghosh et al. 2022), or a focus on 𝑟-process or timescale compared to the nuclear burning timescale, one can
neutrino-driven winds in supernovae (e.g., Qian & Woosley trace the ejecta with passively advected particles whose move-
1996; Cardall & Fuller 1997; Hoffman et al. 1997; Otsuki et al. ments are influenced by the velocity field of the fluid (e.g.,
2000; Thompson et al. 2001; Wanajo et al. 2001; Fröhlich Nagataki et al. 1997; Seitenzahl et al. 2010; Nishimura et al.
et al. 2006a; Kratz et al. 2008; Bliss et al. 2020; Psaltis et al. 2015; Harris et al. 2017; Bovard & Rezzolla 2017; Sieverding
2022), magnetorotational supernovae (Nishimura et al. 2006; et al. 2023). These so-called tracer particles record the ther-
Winteler et al. 2012; Nishimura et al. 2015, 2017b; Mösta modynamic conditions as well as the neutrino fluxes in time.
et al. 2018; Halevi & Mösta 2018; Reichert et al. 2021; Pow- In the case that the impact of diffusion on the composition is
ell et al. 2023; Reichert et al. 2023), collapsars (MacFadyen negligible compared to the burning, each tracer can be calcu-
& Woosley 1999; Pruet et al. 2003; Surman & McLaughlin lated individually, and the total ejected matter is the (possibly
2004; McLaughlin & Surman 2005; Fujimoto et al. 2008; weighted) average over all tracer particles individually. Reac-
Siegel et al. 2019; Miller et al. 2020; Zenati et al. 2020; tion networks that are based on individual tracers (or zones)
Barnes & Metzger 2022; Just et al. 2022a), and neutron star that are unable to interact with each other are called single-
mergers (e.g., Freiburghaus et al. 1999; Korobkin et al. 2012; zone nuclear reaction networks. The advantage of those codes
Martin et al. 2015; Bovard et al. 2017; Lippuner et al. 2017; is that they can include a much more complete set of nuclei
Wu et al. 2016, 2019; Holmbeck et al. 2019; Wanajo et al. and reactions. This enables a calculation of the synthesis of
2021; Rosswog & Korobkin 2022; Kullmann et al. 2022, the heaviest known elements, typically with ∼ 7000 nuclei
2023). Without nucleosynthesis calculations, a whole layer and ∼ 90000 reactions involved.
of information and observables would remain inaccessible. The compilation of a consistent reaction database is espe-
Some applications require a complex modeling that takes cially challenging. Nuclear reactions are often provided in
species diffusion or convective mixing and nuclear burning si- different formats and in different databases that are individu-
multaneously into account (such as, e.g., the oxygen-burning ally complete. Among others, the largest and publicly avail-
phase of a star or the rapid accreting white dwarfs; Hix & able databases are the JINA Reaclib database (Cyburt et al.
Thielemann 1999; Denissenkov et al. 2019) and the nuclear 2010), Bruslib (Aikawa et al. 2005), the Starlib database (Sal-
reaction network must therefore be included in a hydrody- laska et al. 2013), NACRE (Xu et al. 2013), and KADONIS
namical simulation. This often has the consequence that only (Dillmann et al. 2006). However, none of the aforementioned
a restricted number of nuclei are considered in the calcula- libraries provides a complete set of electron/positron-captures
tion (from the 13 or 14 alpha nuclei network - 13 or 14𝛼 – as well as 𝛽+ /𝛽 − decays at stellar conditions (Fuller et al. 1982,
developed by Thielemann and used e.g. in Mueller 1986; 1985; Oda et al. 1994; Langanke & Martı́nez-Pinedo 2001;
Benz et al. 1989; Livne & Arnett 1995; Garcia-Senz et al. Pruet & Fuller 2003; Suzuki et al. 2016), neutrino reactions
2013; Garcı́a-Senz et al. 2016) over small quasi-equilibrium (e.g., Bruenn 1986; Langanke & Kolbe 2002; Fröhlich et al.
networks (e.g. Hix et al. 1998; Timmes et al. 2000; Hix et al. 2006a; Sieverding et al. 2018, 2019), or fission reactions and
2007, named QE-reduced or iso7), to slightly enlarged net- fragment distributions (Panov et al. 2005; Goriely et al. 2009;
works beyond 13𝛼 – like net21 – which include additional Panov et al. 2010; Petermann et al. 2012; Eichler et al. 2015;
neutron-rich isotopes in the Fe-group in order to be able to Vassh et al. 2019). For an almost complete survey of all of
follow 𝑌𝑒 below 0.5 (for a comparison of these approaches, these resources, see the JINAWEB collected list.1 Therefore,
see Bravo 2020). Recently such methods have been extended nuclear reaction networks always have to perform a certain
to networks that contain up to the order of 100 nuclei (Harris amount of merging of the reaction rates if one wants to use a
et al. 2017; Sandoval et al. 2021; Navó et al. 2023). These complete as possible set of reaction rates. Doing this rigor-
so-called in situ networks have the advantage of providing ously can be a major task, as the consistency depends not only
an accurate nuclear energy production as well as more pre- on not adding reactions twice or leaving them out, but also on
cise nucleon abundances that imply more realistic neutrino adding reactions with the same underlying nuclear input, such
opacities for the feedback to the simulation (e.g., Mueller as the same nuclear masses, which, far from stable nuclei, are
1986; Nakamura et al. 2014; Harris et al. 2017; Navó et al. theoretically calculated.
2023). On the other hand, simplifying assumptions within From a numerical point of view, reaction networks can
the nuclear reaction network equations, artificial numerical be challenging as well. The huge differences in timescales
diffusion (e.g., Fryxell et al. 1991; Hix & Thielemann 1999; of the reaction rates (e.g., weak decays versus strong reac-
Plewa & Müller 1999), and the reduced set of nuclei in such tions) introduce a stiffness into the differential equations. As
energy generation networks can make the predicted ejecta
composition, even with extended post-processing networks,
1 https://www.jinaweb.org/science-research/scientific-resources/data
more uncertain (this is nicely shown in Bravo 2020).
 3

a consequence, explicit integration methods become unsta- networks as well as provides insight of the implementations
ble and implicit methods have to be applied. A full implicit within WinNet.
implementation was first achieved by Truran et al. (1966b), In Section 2 we present the fundamental physic concepts
Truran et al. (1967), Arnett & Truran (1969), Woosley et al. for nuclear reaction networks. This includes the derivation
(1973), Arnould (1976), and Thielemann et al. (1979). While of the ordinary differential equation (ODE) that is solved
nowadays usually the first-order implicit Euler scheme is used in WinNet (Section 2.1), the principle of detailed balance
within large nuclear reaction networks, tests with higher-order (Section 2.2), the concept of nuclear statistical equilibrium
implicit schemes such as the Gear scheme have been per- (NSE; Section 2.3), a method to account for nuclear energy
formed as well (e.g., Timmes 1999; Longland et al. 2014). generation within the temperature evolution (Section 2.4),
There exist a variety of single-zone reaction networks with and the treatment of Coulomb corrections (Section 2.5). The
fully implicit schemes in the literature, e.g., the SantaCruz- code structure and included numerical solvers are presented
code by the Woosley group, going back to Woosley et al. in Section 3. The different supported reaction rate formats
(1973), which followed Arnett (1969) and Truran et al. are introduced in Section 4. Applications and test cases are
(1966a) introducing a complete Newton-Raphson scheme, presented in Section 5. We close with a summary in Section 6.
BasNet (Thielemann et al. 1979; for an early comparison of
the two codes and the implemented reaction rate libraries, see 2. NUCLEAR REACTION NETWORK
Hoffman et al. 1999), Xnet (Hix & Thielemann 1999), rNET FUNDAMENTALS
(Wanajo et al. 2001), CFNet (Fröhlich et al. 2006b), NucNet 2.1. Nuclear reaction networks
(Meyer & Adams 2007), rjava (Kostka et al. 2014), Torch
The fundamental theory behind nuclear reaction networks
(Paxton et al. 2015), GSInet (Mendoza-Temis et al. 2015),
reaches back far into the past (see, e.g., Truran et al. 1966a;
SkyNet (Lippuner & Roberts 2017), Prism (Mumpower et al.
Clayton 1968; Arnett & Truran 1969; Woosley et al. 1973;
2018; Sprouse et al. 2021), pynucastro (Smith et al. 2023),
Hix & Thielemann 1999; Hix & Meyer 2006; Winteler et al.
and other unnamed reaction networks (e.g., Timmes 1999; Il-
2012; Lippuner & Roberts 2017). Here we repeat briefly how
iadis et al. 2002; Otsuki et al. 2003; Koike et al. 2004; Goriely
to derive the differential equations, but refer the reader to
et al. 2011). 2 However, only a small subset of them is pub-
previous publications for more details.
licly available, among them pynucastro3, Torch4, rjava5,
The cross section of a reaction
NucNet6, Xnet7, and SkyNet8.
Here we present the single-zone nuclear reaction network number of reactions per target and second
𝜎= , (1)
code WinNet, an updated version of the reaction network that flux of incoming particles
has been first used in the context of Big Bang nucleosynthesis
in Vonlanthen et al. (2009) and later for calculating the syn- is related to the probability of a nucleus 𝑖 to react with nucleus
thesis of heavy elements in Winteler et al. (2012). WinNet 𝑗. If (e.g. in laboratory conditions like accelerator experi-
has a common origin to many other previously mentioned re- ments) the relative velocity between target 𝑖 and projectile 𝑗
action networks such as XNet, CFNet, and GSINet as all of is a constant value 𝑣, it is given by
them were influenced by BasNet, which served as an initial
𝑟/𝑛𝑖
template. 𝜎= . (2)
𝑛𝑗𝑣
WinNet had been already used for different astrophysics
problems; however, it was not publicly available. The code Here, 𝑟 is the number of reactions per volume and time, 𝑛𝑖
has been entirely written in Fortran 90 and has a user-friendly and 𝑛 𝑗 are the number densities of the target and projectile,
interface. WinNet is able to merge reaction rates from mul- respectively. In an astrophysical plasma, both target and pro-
tiple sources and is designed for high-performance computa- jectile follow specific velocity distributions depending on the
tions. It includes two fully implicit schemes, the first-order environmental conditions like temperature and density (and
implicit Euler-backward scheme and the higher-order Gear the reaction cross section is that of a target with thermally pop-
scheme. This paper presents the basics of nuclear reaction ulated excited states; e.g. Fowler 1974; Holmes et al. 1976;
Rauscher & Thielemann 2000; Rauscher 2022). For an arbi-
2 see also https://cococubed.com/code pages/burn.shtml
trary velocity distribution, 𝑟 can be expressed as:
3 https://github.com/pynucastro/pynucastro/ ∫
4 https://cococubed.com/code pages/net torch.shtml 𝑟 𝑖, 𝑗 = 𝜎(|®𝑣𝑖 − ®𝑣 𝑗 |) · |®𝑣𝑖 − ®𝑣 𝑗 |d𝑛𝑖 d𝑛 𝑗 . (3)
5 https://quarknova.ca/rJava/index.html
6 https://sourceforge.net/u/mbradle/blog/ In thermal equilibrium, the velocity (or momentum or energy)
7 https://github.com/starkiller-astro/XNet
distribution depends on the type of particle, i.e., photons
8 https://bitbucket.org/jlippuner/skynet/
obey a Planck distribution and nuclei obey, in most cases,
4

a Maxwell-Boltzmann distribution. Therefore, for photons, where we introduced the factors 𝑁 𝑖𝑗 , 𝑁 𝑖𝑗,𝑘 , 𝑁 𝑖𝑗,𝑘,𝑙 that account
d𝑛 𝛾 is given by for the number of particles 𝑖 that gets destroyed (negative) or
created (positive) in the reaction. The first term in the equa-
1 𝐸 𝛾2 ® (𝑛𝑖 ®𝑣), accounts for changes due to the fluid flow with
𝑑𝑛 𝛾 = d𝐸 𝛾 (4) tion, −∇·
𝜋 2 (𝑐ℏ) 3 exp 𝐸 𝛾 /(𝑘 B𝑇) − 1
 
velocity ®𝑣 and the last term 𝐶nuc accounts for changes due to
and for nuclei d𝑛𝑖 is expressed by nuclear reactions. We can reformulate the previous equation
! by using the Lagrangian time derivative that is related to the
 3/2
𝑚 𝑖 𝑣𝑖2 3

𝑚𝑖 Eulerian time derivative via
d𝑛𝑖 = 𝑛𝑖 exp − d ®𝑣𝑖 ≡ 𝑛𝑖 𝜙(𝑣𝑖 ) d3 ®𝑣𝑖 ,
2𝜋𝑘 B𝑇 2𝑘 B𝑇 𝐷 𝜕
(5) = ®
+ ®𝑣 · ∇. (11)
𝐷𝑡 𝜕𝑡
where 𝑚 𝑖 is the nuclear mass. For reactions between a nucleus
and a photon, 𝑟 is therefore given by We can therefore obtain

𝜎(𝐸 𝛾 )𝐸 𝛾2 𝜕𝑛𝑖 𝐷𝑛𝑖

∫ ∞
𝑛𝑖 = ® 𝑖,
− ®𝑣 · ∇𝑛 (12)
𝑟 𝑖,𝛾 = 2 2 3   d𝐸 𝛾 ≡ 𝑛𝑖 𝜆𝑖,𝛾 (𝑇). 𝜕𝑡 𝐷𝑡
𝜋 𝑐 ℏ 0 exp 𝐸 𝛾 /(𝑘 B𝑇) − 1
(6) and, as a consequence, Eq. (10) becomes
Reactions of this type are called photodisintegrations. For
the case of two nuclei, the 𝑟 is given by 𝐷𝑛𝑖 ® · ®𝑣 + 𝐶nuc .
= −𝑛𝑖 ∇ (13)
∫ 𝐷𝑡
𝑟 𝑖, 𝑗 = 𝑛𝑖 𝑛 𝑗 𝜎(|𝑣𝑖 − 𝑣 𝑗 |) × |𝑣𝑖 − 𝑣 𝑗 |𝜙(𝑣𝑖 )𝜙(𝑣 𝑗 )d3 ®𝑣𝑖 d3 ®𝑣 𝑗 Using the continuity equation and the Lagrangian time deriva-
tive
= 𝑛𝑖 𝑛 𝑗 ⟨𝜎𝑣⟩𝑖, 𝑗 , (7)
𝜕𝜌 ® · (𝜌®𝑣) = −®𝑣 · ∇𝜌 ® · ®𝑣 ⇒ 𝐷 𝜌 = −𝜌 ∇
® − 𝜌∇ ® · ®𝑣, (14)
where ⟨𝜎𝑣⟩𝑖, 𝑗 stands for the product 𝜎𝑣 integrated over ther- = −∇
𝜕𝑡 𝐷𝑡
mal distributions. Furthermore, an additional factor has to be
introduced to avoid double counting of identical project and and therefore
target nuclei. Equation (7) becomes ® · ®𝑣 = − 1 𝐷 𝜌 .
∇ (15)
𝜌 𝐷𝑡
1
𝑟 𝑖, 𝑗 = 𝑛𝑖 𝑛 𝑗 ⟨𝜎𝑣⟩𝑖, 𝑗 (8) Thus, we can insert into Eq. (13) and get
1 + 𝛿𝑖 𝑗
with the delta in the usual sense, i.e., 𝛿𝑖, 𝑗 = 1 for 𝑖 = 𝑗; 𝐷𝑛𝑖 𝑛𝑖 𝐷 𝜌 𝐷 (𝑛𝑖 /𝜌)
− =𝜌 = 𝐶nuc . (16)
otherwise, 𝛿𝑖, 𝑗 = 0. We get 𝐷𝑡 𝜌 𝐷𝑡 𝐷𝑡
1 This derivation has been done previously (see, e.g., Mihalas
𝑟 𝑗,𝑘,𝑙 = 𝑛 𝑗 𝑛 𝑘 𝑛𝑙 ⟨𝑖 𝑗 𝑘⟩
1 + 𝛿 𝑗 𝑘 + 𝛿 𝑘𝑙 + 𝛿 𝑗𝑙 + 2𝛿 𝑗 𝑘𝑙 1999) in the context of atomic processes related to radiation
1 transport, but as shown here it is also valid in the context of
≡ 𝑛 𝑗 𝑛 𝑘 𝑛𝑙 ⟨𝑖 𝑗 𝑘⟩. (9) nuclear reactions (see also Lippuner & Roberts 2017).
1 + Δ 𝑗 𝑘𝑙
In order to obtain a density-independent expression instead
in case of three participating nuclei, where ⟨𝑖 𝑗 𝑘⟩ stands for of utilizing number densities 𝑛𝑖 , we introduce the density (or
three-body reactions (in most cases, a sequence of two two- mass) fraction of nucleus 𝑖, 𝑋𝑖 , which can be expressed via
body reactions and an intermediate reaction product with an
extremely short half-life; see, e.g. Nomoto et al. 1985; Görres 𝜌 𝑖 𝑛 𝑖 𝑚 𝑖 𝑛 𝑖 A 𝑖 𝑚 𝑢 𝑛 𝑖 𝐴𝑖 𝑚 𝑢
𝑋𝑖 = = = ≈ . (17)
et al. 1995). For example, the triple 𝛼 reaction, which de- 𝜌 𝜌 𝜌 𝜌
scribes the probability of three helium nuclei to form 12 C has
a pre-factor of 1/(1 + Δ 𝑗 𝑘𝑙 ) = 1/6. This includes the mass of nuclei A𝑖 𝑚 𝑢 (where A𝑖 is the
Within a fluid that moves with velocity ®𝑣, 𝑛𝑖 does not only relative atomic mass, which can be with a permille error
change by nuclear reactions but also by the net-flow into the approximated by 𝐴𝑖 , the mass number of nucleus 𝑖, and 𝑚 𝑢 =
volume. We have 𝑚( 12 C)/12 is the atomic mass unit). Alternatively, one can
introduce an abundance, without the inclusion of the weight
𝜕𝑛𝑖 ® · (𝑛𝑖 ®𝑣) +
∑︁ ∑︁
=−∇ 𝑁 𝑖𝑗 𝑟 𝑗 + 𝑁 𝑖𝑗,𝑘 𝑟 𝑗,𝑘 or mass of a nucleus, as the fraction of the number density
𝜕𝑡 𝑗 𝑗,𝑘 of nucleus 𝑖 in comparison to the total number density of
nucleons, approximated by 𝑛 = 𝜌/𝑚 𝑢 , that is conserved by
∑︁
𝑖
+ 𝑁 𝑗,𝑘,𝑙 𝑟 𝑗,𝑘,𝑙
𝑗,𝑘,𝑙
nuclear reactions
® · (𝑛𝑖 ®𝑣) + 𝐶nuc ,
B−∇ (10) 𝑛𝑖 𝑛𝑖 𝑋𝑖 𝑛𝑖
𝑌𝑖 = = = = 𝑁A 𝑚 𝑢 . (18)
𝑛 𝜌/𝑚 𝑢 𝐴𝑖 𝜌𝑁A
 5

This definition seems to differ from the traditionally utilized defined as the difference between initial and final ground-state
one in nucleosynthesis literature, introduced by Fowler et al. masses. The relation between both can be expressed as (e.g.,
(1967) Fowler et al. 1967)
𝑛𝑖
𝑌𝑖 = , (19) Î 𝑁reactants
𝜌𝑁A Δreactants © 𝑖=1 𝐺 𝑖 (𝑇) ª
⟨𝜎𝑣⟩backward = ­ Î 𝑁products ® (21)
as it includes the Avogadro constant 𝑁A rather than the nuclear Δproducts 𝐺 𝑗 (𝑇)
𝑗=1
mass unit 𝑚 𝑢 . Eqs. (19) and (18) differ by the product 𝑀𝑢 =
« ¬
Î 𝑁reactants Î 𝑁reactants 3𝑛/2
𝑁A 𝑚 𝑢 , the molar gas constant, which had until 2019 the value © 𝑖=1 𝑔𝑖 ª © 𝑖=1 𝐴𝑖 ª
10−3 kg/mole or 1 g/mole in cgs units, leading in Eq. (19) × ­ Î𝑁 ® ­ Î 𝑁products ®
products
𝑔 𝑗 𝐴 𝑗
to an abundances measure in mole/g and in Eq. (18) to a « 𝑗=1 ¬ « 𝑗=1 ¬
  3𝑛/2
dimensionless number. 𝑚 𝑢 𝑘 B𝑇
When utilizing the present values of the natural con- × exp [−𝑄/(𝑘 B𝑇)] ⟨𝜎𝑣⟩forward ,
2𝜋ℏ2
stants (see Table XXXI in Tiesinga et al. 2021) with 𝑁A =
where Δ is the double counting factor for reactants/products
6.02214076 × 1023 mole−1 (exact) and 𝑚 𝑢 = 𝑚( 12 C)/12 =
as in Eq. (20), 𝐺 are the partition functions, 𝑔 is the spin factor
1.6605390660(50) × 10−24 g with a relative uncertainty of
defined as 𝑔 = 2𝐽 + 1, where 𝐽 is the spin of the ground state,
3 × 10−10 , one obtains for the molar mass constant 𝑀𝑢 =
𝐴𝑖 is the mass of nucleus 𝑖, 𝑄 is the Q-value of the reaction,
𝑁A 𝑚 𝑢 = 0.99999999965(30) g mol−1 , i.e. equal to 1 with
⟨𝜎𝑣⟩forward is the cross section of the forward reaction, and
an uncertainty of 3 × 10−10 . Thus, both expressions are nu-
𝑛 is the difference between the number of reactants and the
merically identical with an extremely high accuracy in cgs
number of reaction products.
units. However, the different dimensions of Eqs. (18) and
Equation (21) needs to be modified for photodisintegration
(19) can introduce some confusion (see also Rauscher 2020).
reactions. There, ⟨𝜎𝜈⟩backward should be replaced by 𝜆backward .
In this paper, we continue to utilize the traditional definition
In this case, 𝑛 ≠ 0 and we therefore get the additional factors
for abundances (Eq. (19), but in agreement with Eq. (17) and
that are introduced with 𝑛 in the exponential in Eq. (21).
𝑌𝑖 = 𝑋𝑖 /𝐴𝑖 we will treat mass fractions 𝑋𝑖 as well as abun-
This is consistent with literature (e.g., Fowler et al. 1967)
dances 𝑌𝑖 as dimensionless numbers. When expressing the
and the Reaclib reverse rates. Therefore, in practice, we can
number densities 𝑛𝑖 in terms of abundances 𝑌𝑖 , Eq. (16) leads
use the above equation for both cases, capture reactions and
to the form
photodisintegrations. Eq. (21) is also valid for three-body
𝐷𝑌𝑖 ∑︁ 𝑖 reactions replacing ⟨𝜎𝑣⟩ by ⟨𝑖 𝑗 𝑘⟩. It should be mentioned
𝑌¤𝑖 = = 𝑁 𝑗 𝜆 𝑗𝑌 𝑗 (1-body)
𝐷𝑡 𝑗
here that the relations in this section include that nuclei in a
thermal environment exist with thermally populated excited
∑︁ 𝑁 𝑖𝑗,𝑘
+ 𝜌𝑁A ⟨𝜎𝑣⟩ 𝑗,𝑘 𝑌 𝑗 𝑌𝑘 (2-body) states.
𝑗,𝑘
1 + 𝛿 𝑗𝑘 Within the Jina Reaclib framework, the Q-Values are given
∑︁ 𝑁 𝑖𝑗,𝑘,𝑙 for each reaction. Additionally, the mass excesses of all nu-
+ 𝜌 2 𝑁A2 ⟨𝑖 𝑗 𝑘⟩𝑌 𝑗 𝑌𝑘 𝑌𝑙 . (3-body), clei can be found in a separate file (called ”winvn”). Ideally,
𝑗,𝑘,𝑙
1 + Δ 𝑗 𝑘𝑙 the mass excess is consistent with the Q-value in the Rea-
(20) clib; however, as pointed out already in Lippuner & Roberts
(2017), currently there are inconsistencies between these val-
where the individual terms can be identified with specific ues. Because the reverse rates in Reaclib use the detailed bal-
reactions, neglecting reactions involving four or more partici- ance principle with the Q-value from the Reaclib, there can
pants. The first term, standing for one-body reactions, usually be an inconsistency at the transition of NSE to the network
includes decays, photodisintegrations, electron- or positron- equations caused by the inconsistent Q-values (see Fig. 2).
captures, and neutrino absorption. The equation is often Therefore, WinNet is able to calculate detailed balance with
called the ”nuclear reaction network equation”. It is the fun- the Q-values obtained from the mass excess. We note that
damental differential equation that is solved within WinNet. there is no optimal solution for this inconsistency. Using the
Note that all 𝜌𝑁A terms would be replaced by 𝜌/𝑚 𝑢 , when Q-value from the mass excess will make the calculation con-
utilizing the alternative definition of abundances 𝑌𝑖 , which sistent with NSE, but introduces an inconsistency with the
would replace 𝑁A by 𝑚 𝑢−1 . forward rate, as this was calculated on the basis of a different
Q-value. Often, it is however more important to be consis-
2.2. Detailed balance tent with the equilibrium values. As already mentioned in
Reverse or backward reactions have a direct relation to the Lippuner & Roberts (2017), this inconsistency in the Reaclib
forward reaction by the so-called detailed balance theorem. database may be resolved in the future. However, one phi-
We denote as forward reaction those with a positive Q-value losophy of the JINA Reaclib database is to have up-to-date
6

nuclear masses. Recalculating all reaction rates whenever Within WinNet we solve the system of Equations (24)
a new mass is available may not be feasible. To a certain and (25) either with a Newton-Raphson or identical to Smith
degree, this inconsistency may therefore always be present et al. (2023) the hybrid Powell method from the MINPACK-
(Schatz 2022). In any case, the advantage of an on-the-fly I package (More et al. 1980) that was translated to Fortran
calculation of reverse rates is also given when using tabulated 90 by J. Burkardt9. The convergence of the schemes hereby
rates. For these rates, a tabulation for forward and reverse depend often on the initial guess. In WinNet this guess is
rates may break the detailed balance principle, and it can obtained by starting to calculate NSE at a high temperature
be more consistent to calculate the reverse rate based on the and descending to lower temperatures, taking the results of the
tabulation of the forward rate. higher temperatures as initial value for the lower ones. The
initial composition at the starting temperature is assumed to
2.3. Nuclear statistical equilibrium consist of nucleons only with 𝑌𝑛 = 1 − 𝑌𝑒 and 𝑌 𝑝 = 𝑌𝑒 .
For high temperatures in explosive environments, typically Weak reactions are evolved with a simplified reaction net-
in excess of about 𝑇 ≳ 6 GK, reactions mediated by the work that includes only these reactions in Eq. (20). After a
strong and electromagnetic interaction are in equilibrium. For time step a new electron fraction is determined using Eq. (25)
these conditions, one can simplify the treatment, replacing and the composition is recomputed assuming for NSE. This
the reaction network equations by utilizing an equilibrium assumes that strong and electromagnetic reactions occur in-
approach, which can be expressed in terms of the chemical stantaneously following a weak reaction consistently with the
potentials of the nuclei NSE assumption.
The implementation of screening corrections in NSE is
𝜇(𝑍, 𝑁) = 𝑁 𝜇 𝑛 + 𝑍 𝜇 𝑝 , (22) discussed in Section 2.5.

where 𝜇(𝑍, 𝑁) is the chemical potential for a nucleus with 2.4. Nuclear heating
mass number 𝐴 = 𝑍 + 𝑁, 𝜇 𝑛 is the chemical potential of A proper consideration of the impact of the energy pro-
neutrons, and 𝜇 𝑝 is the chemical potential of protons. For low duced by nuclear processes in the hydrodynamical evolution
enough densities, nucleons (fermions) are nondegenerate and requires the use of an in situ network as discussed in the
therefore described well by the Maxwell-Boltzmann statistics. introduction. However, in post-processing network calcula-
Introducing this for the related chemical potentials in Eq. (22) tions, it is commonly assumed that the nuclear energy gen-
leads to the so-called Saha equations (for a detailed derivation eration mainly affects the evolution of temperature (see, e.g.
of NSE, see, e.g. Hix & Thielemann 1999; Iliadis 2015; Freiburghaus et al. 1999; Mueller 1986; Lippuner & Roberts
Lippuner & Roberts 2017, or for an approach using detailed 2017). In the following, we describe the general description
balance, see, e.g., Clayton 1968): of energy generation and its treatment in WinNet.
The evolution of a fluid element under exchange of heat
 3 ( 𝐴−1)
𝐴3/2 with the surroundings in a local inertial frame comoving with

2𝜋ℏ 2
𝑌 (𝑍, 𝐴) = 𝑔 𝑍 , 𝐴𝐺 𝑍 , 𝐴 (𝜌𝑁 𝐴) 𝐴−1 the fluid is given by the first law of thermodynamics
2 𝐴 𝑚 𝑢 𝑘 B𝑇
𝑒 𝐵𝑍, 𝐴/𝑘𝐵 𝑇 𝑌𝑛𝐴−𝑍 𝑌 𝑝𝑍 , (23)  
1
𝑑𝜀 + 𝑝𝑑 = 𝛿𝑞 , (26)
𝑛
with the spin factor 𝑔 = 2𝐽 + 1, where 𝐽 is the spin of the
ground state, the partition function 𝐺 𝑍 , 𝐴, and binding energy where 𝜀 is the total energy (including rest-mass energy) per
of a nucleus 𝐵 𝑍 , 𝐴. Furthermore, additional constraints of nucleon, and 𝛿𝑞 is the net heat gained per nucleon. This
mass conservation and charge neutrality hold: includes heat produced by shocks or viscous heating or loss by
∑︁ neutrinos when weak processes are considered. Alternatively,
𝑌𝑖 𝐴𝑖 = 1 (mass conservation) (24) if the fluid element is in equilibrium at all times, we have
𝑖
∑︁ ∑︁
𝑌𝑖 𝑍𝑖 = 𝑌𝑒 (charge neutrality). (25) 𝑘 𝐵𝑇 𝑑𝑠 + 𝜇𝑖 𝑑𝑌𝑖 + 𝜇𝑒 𝑑𝑌𝑒 = 𝛿𝑞 (27)
𝑖 𝑖
∑︁
This set of equations has two unknowns, namely the abun- = 𝑘 𝐵𝑇 𝑑𝑠 + (𝜇𝑖 + 𝑍𝑖 𝜇𝑒 ) 𝑑𝑌𝑖 ,
dances of protons 𝑌 𝑝 and neutrons 𝑌𝑛 , because temperature, 𝑖
density, and electron fraction are assumed to be known quan- where 𝑠 is the entropy per nucleon in units of 𝑘 𝐵 and the sum
tities (e.g., from a hydrodynamical simulation). The com- runs over all nuclear species. The term 𝜇𝑒 𝑑𝑌𝑒 or 𝑍𝑖 𝜇𝑒 𝑑𝑌𝑖
position is a function of 𝑌 (𝜌, 𝑇, 𝑌𝑒 ) only. In particular, no
information of the past behavior is necessary to determine the
composition. 9 Accessed from https://people.sc.fsu.edu/∼jburkardt/f src/fsolve/fsolve.html
 7

accounts for the contribution of electrons and positrons. Typ- implementation of neutrino reactions). For the moment, we
ically, the densities we are interested in are such that matter is include 𝜀 𝜈 for charged-current reactions on nucleons only.
transparent to neutrinos. To ensure this, within WinNet we When combining Eq. (27), (30), and (31), we obtain:
include a user-defined parameter to specify the density below !
which nuclear heating will be taken into account. The energy 1 ∑︁
𝑠¤ = − (𝜇𝑖 + 𝑍𝑖 𝜇𝑒 ) 𝑌¤𝑖 − 𝑞¤
carried by neutrinos per unit of time can be expressed as: 𝑘 𝐵𝑇 𝑖
∑︁ ! (32)
𝑞¤ loss = − ⟨𝜀 𝜈 ⟩ 𝑖 𝜆𝑖𝑌𝑖 , (28) 1 ∑︁ ¤
=− (𝜇𝑖 + 𝑍𝑖 𝜇𝑒 ) 𝑌𝑖 − ( 𝑞¤ loss + 𝑞¤ gain ) ,
𝑖 𝑘 𝐵𝑇 𝑖
where ⟨𝜀 𝜈 is the average energy of the neutrinos produced
⟩𝑖
by electron-capture or beta-decay of the nucleus 𝑖 with rate where we obtain the electron chemical potential from the
𝜆𝑖 and abundance 𝑌𝑖 . These quantities are provided in tab- EOS (Timmes & Arnett 1999), and the chemical potentials
ulations of weak interaction rates at finite temperature and of nuclei is given by:
density (see, e.g., Langanke & Martı́nez-Pinedo 2001) and in
𝜇𝑖 = 𝑚 𝑖 𝑐2 + 𝑘 𝐵𝑇𝜂𝑖
global calculations of beta-decays for 𝑟-process nuclei (Mar- "  3/2 #
𝑔𝑖 𝐺 𝑖 (𝑇)𝑚 𝑢 𝐴𝑖 𝑚 𝑢 𝑐2 𝑘 𝐵𝑇 (33)

ketin et al. 2016). For measured decays, the average neutrino
𝜂𝑖 = − ln .
energies are given by the ENSDF database (Brown et al. 𝜌𝑌𝑖 2𝜋ℏ2
2018)10. If we consider only beta-decays, we can express
the average energy ⟨𝜀 𝜈 ⟩ 𝑖 of the neutrinos as a fraction of the Here 𝑚 𝑖 is the nuclear mass that we get from the atomic mass
beta-decay Q-value 𝑄 𝛽,𝑖 : excess Δ(𝑍, 𝐴) by
∑︁
𝑞¤ loss = − 𝑓 𝜈,𝑖 𝑄 𝛽,𝑖 𝜆𝑖𝑌𝑖 . (29) 𝑚(𝑍, 𝐴)𝑐2 = Δ(𝑍, 𝐴) + 𝐴𝑚 𝑢 𝑐2 − 𝑍𝑚 𝑒 𝑐2 . (34)
𝑖
The mass excess from the latest atomic mass evaluation is
Assuming that a constant fraction of the energy is carried by
tabulated in the Jina Reaclib database (Cyburt et al. 2010).
neutrinos, we have
Under NSE conditions, Eq. (32) can be expressed as
∑︁
𝑞¤ loss = − 𝑓 𝜈 𝑄 𝛽,𝑖 𝜆𝑖𝑌𝑖 . (30)
1 
(𝜇 𝑝 + 𝜇𝑒 − 𝜇 𝑛 )𝑌¤𝑒 − 𝑞¤ ,

𝑖 𝑠¤ = − (35)
𝑘 𝐵𝑇
A typical value of 𝑓 𝜈 for neutron-rich 𝑟-process nuclei is
𝑓 𝜈 = 0.4 (Marketin et al. 2016), as beta-decays populate showing that only reactions that are not in equilibrium, i.e.
mainly excited states in the daughter nuclei that later decay weak processes that change 𝑌𝑒 as well as external heating, are
by either 𝛾 or neutron emission. In practice, within WinNet responsible for the change in entropy. This result can be also
the average energy of neutrinos produced in the reaction can generalized to 𝑟-process conditions for which (𝑛, 𝛾) ⇄ (𝛾, 𝑛)
be taken from all aforementioned sources, and in case of an equilibrium is valid. Hence, reactions in equilibrium do not
unknown average neutrino energy, a user-defined 𝑓 𝜈 is as- introduce a change in entropy. At high densities, neutrinos are
sumed. Optionally, we also account for escaping thermally characterized by a chemical potential 𝜇 𝜈 . In this case, one
produced neutrinos, by, e.g., bremsstrahlung or electron re- obtains 𝑠¤ = − (𝜇 𝑝 + 𝜇𝑒 − 𝜇 𝑛 − 𝜇 𝜈 )𝑌¤𝑒 − 𝑞¤ /(𝑘 𝐵𝑇), which
combination with the analytic fitting formulas of Itoh et al. shows chemical weak equilibrium 𝜇 𝑝 + 𝜇𝑒 = 𝜇 𝑛 + 𝜇 𝜈 corre-
(1996)11. Energy cannot only leave the system by neutrinos, sponds to a maximum of the entropy (Arcones et al. 2010).
but also enter it. When assuming that only neutrino reactions Within WinNet we solve Eq. (32) explicitly in a so-called
add additional energy to the system, we obtain: operator splitting method within the same Newton-Raphson
∑︁ as the nuclear network equations (Eq. 20). The initial value
𝑞¤ gain = ⟨𝜀 𝜈 ⟩ 𝑖 ⟨𝜎⟩𝑖 𝐹𝜈𝑌𝑖 , (31) of the entropy is determined using the Timmes EOS (Timmes
𝑖 & Arnett 1999). Within every Newton-Raphson iteration, the
newly obtained entropy is translated into a temperature via
where ⟨𝜀 𝜈 ⟩ is the average energy of the absorbed neutrino,
the EOS assuming that the density and composition remain
⟨𝜎⟩ is the neutrino average cross section, and 𝐹𝜈 the neutrino
constant. For conditions at which 𝑠 ≈ 1–5 𝑘 B nuc−1 , the
number flux (see Section 4.2.4 for more details about the
entropy is dominated by the contribution of nuclei and is
very sensitive to the composition. Under these conditions, it
10 Accessed via the API of https://www-nds.iaea.org/relnsd/vcharthtml/api is necessary to account for changes in the composition when
v0 guide.html searching for a new value of the temperature. This is currently
11 See also https://cococubed.com/code pages/nuloss.shtml not implemented in WinNet.
8

2.5. Coulomb corrections Kravchuk & Yakovlev (2014) took from Potekhin & Chabrier
Coulomb effects can significantly influence fusion pro- (2000):
cesses in a hot stellar plasma. Electrons can be attracted
" √︂ √︂ !#
√︁ Γ Γ
by the positive charge of a nucleus and therefore shield and 𝑓𝐶 (Γ) = 𝐶1 Γ(𝐶2 + Γ) − 𝐶2 ln + 1+
𝐶2 𝐶2
modify the Coulomb interactions between two nuclei. This √ √ 
modifies the nuclear reactions and makes charged particle + 2𝐶3 Γ − arctan Γ
reactions more likely. The effect can be approximated by cor-   
Γ
rection factors, the so-called screening corrections, which are + 𝐷 1 Γ − 𝐷 2 ln 1 +
an important ingredient in nuclear reaction network calcula- 𝐷2
2
 
tions (e.g., Salpeter 1954). The calculation of the correction 𝐷3 Γ
+ ln 1 + . (40)
factors depends on the temperature and density of the envi- 2 𝐷4
ronment (e.g., Salpeter & van Horn 1969; Yakovlev & Shaly- Here, 𝐶1 = −0.907, 𝐶2 = 0.62954, 𝐶3 = 0.2771, 𝐷 1 =
bkov 1989; Ichimaru 1993; Yakovlev et al. 2006). Usually, 0.00456, 𝐷 2 = 211.6, 𝐷 3 = −0.0001, and 𝐷 4 = 0.00462,
three different screening regimes are distinguished: the weak and Γ is the ion-coupling parameter for a one-component
screening, the intermediate screening, and the strong screen- plasma
ing regime. The regimes are commonly separated in terms 𝑍 5/3 𝑒 2 (4𝜋𝑛𝑒 ) 1/3
of the ion-coupling parameter (e.g., Kravchuk & Yakovlev Γ= . (41)
31/3 𝑘 B𝑇
2014)
From this, they obtain
𝑍1 𝑍2 𝑒 2 (4𝜋𝑛𝑒 ) 1/3 𝑓𝐶 (Γ1 ) + 𝑓𝐶 (Γ2 ) − 𝑓𝐶 (Γ𝐶 )
Γ12 = 2 𝑏0 = , (42)
𝑍11/3 + 𝑍21/3 31/3 𝑘 B𝑇 Γ12
𝑍1 𝑍2 where Γ1 and Γ2 are the ion-coupling parameters of the re-
≈ 4.5494 × 10−4 (𝜌𝑌e ) 1/3𝑇 −1 , (36)
𝑍11/3 + 𝑍21/3 acting nuclei, and Γ𝐶 is the ion coupling parameter of the
compound nucleus. Furthermore, 𝑏 2 and 𝑏 4 in eq. (37) are
where 𝑇 is the temperature in GK, the electron number den- defined as
sity is defined as 𝑛𝑒 = 𝜌𝑁A𝑌𝑒 , Z𝑖 is the charge of element
3
𝑖, and the elementary charge is 𝑒. For lower values of Γ12 1 1 + 𝑧5/3
𝑏2 = − (43)
the effect of screening becomes smaller. The weak screening 16 1 + 𝑧
5
regime applies for Γ12 ≪ 1, the intermediate regime around 𝑧 1 + 𝑧5/3


Γ12 ≈ 1, and the strong regime for larger values. We do not 𝑏4 = − . (44)
64 (1 + 𝑧) 11/3
solve the screening corrections numerically, which would be
necessary to obtain the corrections for the strong screening The differences between the screening correction scheme of
regime. Instead, we have implemented a fitted function that Kravchuk & Yakovlev (2014) that is implemented in WinNet
was derived within Kravchuk & Yakovlev (2014). They ex- and that of SkyNet (Lippuner & Roberts 2017), which uses
press the so-called screening enhancement factor as (Eq. 62 a parameterization of Dewitt et al. (1973) are shown in Fig. 1.
of Kravchuk & Yakovlev 2014) In the most relevant regime for nucleosynthesis calculations
   (i.e., 1 ≤ Γ12 ≤ 200) all schemes show good agreement
5 63 (Fig. 1). For higher values of Γ12 > 200, the temperature is
𝑓scr = exp Γ12 𝑏 0 + 𝑏 2 𝜁 2 + 𝑏4 𝜁 4 , (37)
8 128 usually close to or even below the validity of the reaction rate
databases (c.f., 𝑇min = 10−2 GK of the Reaclib reaction rate
with 𝜁 defined as
Γ12 database; Cyburt et al. 2010).
𝜁 =3 , (38) Screening corrections will modify the reaction rates ac-
𝜏
cording to
where
⟨𝜎𝜈⟩scr = ⟨𝜎𝜈⟩ 𝑓scr . (45)
 1/3
27𝜋 2 𝜇(𝑍1 𝑍2 ) 2 𝑒 4

𝜏= The implementation of screening with more than two reac-
2𝑘 B𝑇ℏ2 tants is realized in several steps. For three reactants, the
  1/3 screening correction of only two reactants is calculated, and,
𝐴1 𝐴2 21
≈ 4.2487 × (𝑍1 𝑍2 ) , (39) in a next step, the correction of the third reactant with the
𝐴1 + 𝐴2 𝑇
summed mass and ion number of the first two reactants is
with the nucleon number 𝐴𝑖 and the reduced mass 𝜇. The calculated. This corresponds to forming a short-lived inter-
fitting parameter 𝑏 0 is expressed by the difference in Coulomb mediate nucleus. The total correction is then given by the
free energies, which are defined by another fitted function that multiplication of both correction factors 𝑓scr .
 9

100
WinNet NSE
80 Increasing 12 SkyNet Network
exp( 12b0)
10 3
Network (det. bal.)
60

12 = 1

Mass fraction X
Log(fScr)

40 10 4
Weak
20 Strong screening
screening regime
regime 12 1
0 10 5
1 Screening
Log(fScr, 1)
Log(fScr, 2)

No screening
0 10 6
25 30 35 40 45 50 55 60 65
10 10 100 101
1

2 1
Mass number A
Temperature [GK]

Figure 2. NSE composition with (orange line) and without (blue

Figure 1. Upper panel: screening correction for the heavy ion line) screening corrections in NSE for 𝑇 = 7 GK, 𝜌 = 107 g cm −3 ,
reaction 12 C+12 C for a constant density of 108 g cm −3 and 𝑌e = 0.5. and 𝑌𝑒 = 0.5 (solid lines). Dashed lines show the result of two
The screening correction of Kravchuk & Yakovlev (2014) that is hydrostatic network runs with and without screening corrections
used in WinNet is shown as the solid green line. The screening using strong reaction rates from the Jina Reaclib. Dashed lines show
correction of Kravchuk & Yakovlev (2014) when only using the 𝑏 0 the same, but replacing the reverse reactions of the Jina Reaclib with
term that is similar to the original description of Salpeter (1954) reverse rates that are calculated with detailed balance using the mass
is shown as the dotted orange line. The screening correction of excess of the Jina Reaclib. The hydrostatic calculations start with
SkyNet for a pure carbon composition is shown as the dashed red half neutrons and half protons and are calculated for 103 s. This
line. Bottom panel: relative differences of the screening corrections illustrates the consistency of the network at NSE transition with and
relative to that implemented in WinNet. The vertical dashed line without screening corrections when using the same nuclear masses
indicates the intermediate screening regime for Γ12 = 1. for the reactions and NSE.

In the case of NSE (see Section 2.3), screening corrections the initialization step. This file contains runtime parameters
enter in the form of a change of the binding energy of a charged such as paths to nuclear physics input data and other options.
nucleus, i.e., the difference in the Helmholtz free energy due A full list of possible parameters is given in the documentation
to the screening. Since all reactions are in equilibrium, we of the code.
can assume that every nucleus is built by a series of proton After the initialization, the evolution mode is chosen. This
captures and neutron captures, where the latter reaction is mode is set to either “Network” or “NSE” and depends on
independent of screening corrections. To obtain a correction the temperature. The implementation of several modes is
for the binding energy of a given nucleus with charge number necessary as the most efficient approach to determine the
𝑍 we therefore multiply the screening corrections 𝑓scr of the composition changes with temperature. Whereas solving the
necessary amount of (𝑍 − 1) proton captures12. The impact of full network equations in a temperature regime where an equi-
screening and the consistency of the network at NSE transi- librium holds can lead to arbitrarily small time steps, solving
tion is shown in Fig. 2. Note that there exist other approaches NSE conditions at too low temperatures can lead to incorrect
that derive the screening corrections from the detailed bal- results.
ance principle (e.g., Kushnir et al. 2019) or from a global For both evolution modes, the temperature, density, and
Coulomb correction (Bravo & Garcı́a-Senz 1999; Lippuner neutrino quantities (i.e., neutrino temperatures or energies
& Roberts 2017). All of these approaches are consistent with and luminosities) are updated using either an interpolation
each other. When taking screening corrections into account, (i.e., linear, cubic, Akima, modified Akima, Pchip) within
heavier nuclei are synthesized compared to the case without the thermodynamic data of the Lagrangian tracer particle,
screening. analytic equations, or a user-defined extrapolation (i.e., adi-
abatic, exponential, free). In the network regime, updating
3. METHODS AND NUMERICAL TECHNIQUES
the temperature depends on the input settings and includes
3.1. Code structure and flow diagram some special cases. If the user allows feedback of the nuclear
In the following, we describe the control flow of WinNet energy release on the temperature, a differential equation of
(see Fig. 3). The code starts by reading a user-defined file in the entropy is solved explicitly together with the nuclear re-
action network equations (see Section 2.4). After updating
the temperature, density, and neutrino properties, the reaction
12 Note that XNet uses the same approach in NSE; see https://github.com/ network equations are solved numerically. For the network
starkiller-astro/XNet/blob/master/doc/screening/Screening for NSE.pdf
10

regime, the full set of coupled differential equations (includ- be numerically unstable for so-called stiff problems that are
ing all reactions) is solved. In NSE, Eqs. (23-25) are solved present in reaction networks. Therefore, we can discretize the
for a given temperature, density, and electron fraction. The derivative with
latter is evolved taking weak reactions into account only.
𝑌𝑖 (𝑡 + ℎ) − 𝑌𝑖 (𝑡)
If no convergence is achieved (the criteria are introduced = 𝑓𝑖 (𝑡 + ℎ, 𝑌1 , ..., 𝑌𝑁 ). (48)
ℎ
in Section 3.2), the step size is halved, and the iteration is
repeated. Otherwise, an output is generated, and the time Note that here 𝑌𝑖 (𝑡 + ℎ) as well as 𝑓𝑖 (𝑡 + ℎ, 𝑌1 , ..., 𝑌𝑁 ) is
is evolved (indicated by “rotate timelevels” in Fig. 3). The unknown. We can derive a iterative formula for the solution
main loop ends when a user-defined termination criterion is of 𝑌𝑖 (𝑡 + ℎ). This is given by:
fulfilled. Before the code terminates, final output such as the
final abundances and mass fractions are written. 𝑌𝑖 (𝑡 + ℎ) = 𝑌𝑖 (𝑡) + ℎ 𝑓𝑖 (𝑡 + ℎ, 𝑌1 , ..., 𝑌𝑁 ). (49)

3.2. Integration schemes To get a solution, we have to apply a root-finding algorithm; in

WinNet we use the Newton-Raphson method. To apply the
Due to the stiff behavior of the nuclear reaction network Newton-Raphson method, we must reformulate the problem:
equations (Eq. (20)), implicit/backwards methods are nec-
essary to integrate the ODE. The general structure of the 0 = 𝑌𝑖 (𝑡) + ℎ 𝑓𝑖 (𝑡 + ℎ, 𝑌1 , ..., 𝑌𝑁 ) − 𝑌𝑖 (𝑡 + ℎ). (50)
network however is independent of the chosen integration
method. Mathematically, a multidimensional Newton-Raphson can be
Regardless of the chosen integration method, WinNet uses formulated as:
a sparse matrix representation of the Jacobian of the system 𝐹® ( 𝑥®) = 0, (51)
and the sparse matrix solver PARDISO (Schenk & Gärtner which we will later apply and set 𝑥® to 𝑌® . The Taylor series of
2004), which is OpenMP parallelized. For a detailed descrip- 𝐹® can be expressed in first order as
tion of the sparse format, see e.g., Hix & Thielemann (1999)
or Winteler (2012). This sparse format brings a computa- 𝑁
∑︁ 𝜕𝐹𝑖
tional advantage for calculations with more than 𝑁 ≳ 400 𝐹𝑖 ( 𝑥® + 𝛿®
𝑥 ) = 𝐹𝑖 ( 𝑥®) + 𝑥 2 ) = 0,
𝛿𝑥 𝑗 + 𝑂 (𝛿® (52)
𝑗=1
𝜕𝑥 𝑗
nuclei. In WinNet, the indices of possible nonvanishing en-
tries are calculated once at the beginning and are updated in 𝜕𝐹𝑖
where is one entry of the Jacobian matrix containing the
a next step when solving the linear system. WinNet pro- 𝜕𝑥 𝑗

vides two methods to integrate the system that are outlined ® defined as 𝐽𝑖 𝑗 = 𝜕𝐹𝑖 . Within WinNet
partial derivatives of 𝐹, 𝜕𝑥 𝑗
subsequently. this Jacobian is represented in a sparse format. The position
of zero entries is reevaluated in every iteration. Furthermore,
3.2.1. Implicit Euler derivatives of the screening correction factor (Section 2.5) are
The implicit Euler method (see also, e.g. Hix & Thiele- considered to be zero, and we calculate the Jacobian analyti-
mann 1999; Winteler 2012; Lippuner & Roberts 2017) is one cally as, in this case, it is just the derivative of a polynomial
of the simplest implicit integration methods. Nevertheless, it equation (Eq. 20). To find the root of 𝐹® we iterate
is sufficient for most of the calculations, especially when a
𝑥 𝑘 = 𝑥®𝑘 − 𝐽 ( 𝑥®𝑘 ) −1 · 𝐹® ( 𝑥®𝑘 )
𝑥®𝑘+1 = 𝑥®𝑘 + 𝛿® (53)
large number of nuclei is involved in the calculation. For a
coupled ODE, we can formulate the problem of integrating until convergence is reached. In a classical Newton-Raphson
the equation by the general form of the convergence criterion is given by | 𝑥®𝑘+1 − 𝑥®𝑘 | < 𝜖NR . In
D𝑌𝑖 WinNet we implemented a different criterion that is based on
= 𝑌¤𝑖 = 𝑓𝑖 (𝑡, 𝑌1 , ..., 𝑌𝑁 ), (46) mass conservation by using the mass fraction 𝑋𝑖 (see Eq. 24):
D𝑡
𝑁 𝑁
where N is the amount of involved species, and 𝑌𝑖 is the abun- ∑︁ ∑︁
dance of species 𝑖. There are two possibilities to discretize 𝑌𝑖 𝐴𝑖 − 1 = 𝑋𝑖 − 1 < 𝜖NR (54)
𝑖=1 𝑖=1
this derivative. The simplest approach would be
where 𝜖NR < 10−5 is used per default in WinNet. As in-
𝑌𝑖 (𝑡 + ℎ) − 𝑌𝑖 (𝑡)
= 𝑓𝑖 (𝑡, 𝑌1 , ..., 𝑌𝑁 ). (47) vestigated by Lippuner & Roberts (2017), this convergence
ℎ
criterion is sufficient for most of the nucleosynthesis calcula-
When choosing a time step ℎ, everything except 𝑌𝑖 (𝑡 + ℎ) tions. Other convergence criteria such as | 𝑥®𝑘+1 −®
𝑥 𝑘 | < 𝜖NR are
is known, and one can integrate the ODE when knowing often too strict and slow down the calculation significantly.
an initial value of 𝑌𝑖 . However, this approach corresponds If the Newton-Raphson does not converge, the calculation is
to an explicit Euler method, an integration scheme that can repeated with a halved time step. This is schematically shown
 11

Initialization
Read User input
(e.g. path to reaction rates)

Read Data
(e.g. reaction rates)

Set
evolution
mode

Network NSE

Estimate timestep

Update dens. and Update temp., dens.

neutrino quantities and neutrino quantities

Yes Nuclear No
heating?

Update temperature

Solve reaction network

Solve reaction
Rotate timelevels equation together Half the timestep Calc. weak reactions
with temperature network equation

Converged?
No No Converged?

Yes Yes

Calc. NSE composition

Not fulfilled
Termination Write iterative output Analyse result
criterion (e.g., heating rate) (e.g., reaction timescales)

Fulfilled

Write final output

(e.g., final abundances)

Terminate

Figure 3. Flow diagram of WinNet. Figure taken from Reichert (2021).

12

by the loop in Fig. 3. In the case of nuclear heating being Timmes 1999, for a discussion of the advantages of higher-
enabled (Section 2.4), the temperature change relative to the order solvers for nuclear reaction networks). In the following,
last Newton-Raphson iteration can also be limited in order we will denote the highest included order with 𝑞. It is a so-
to assure the convergence of the entropy update. We tested called predictor-corrector method, where in a first step, a
the convergence in more detail in Appendix A. By combining rough solution is guessed, and in a second step, this solution
Eq. (53) and Eq. (49), we obtain is corrected until a given precision is reached. The first pre-
! −1 ! diction is based on information of the past behavior of the
1 𝜕 𝑓 (𝑌®𝑛+1
𝑘 ) 𝑌®𝑛+1
𝑘 −𝑌®𝑛 system. Therefore, the so-called Nordsieck vector
𝑌®𝑛+1
𝑘+1
= 𝑌®𝑛+1
𝑘
− ×1− · ® ® 𝑘
− 𝑓 (𝑌𝑛+1 ) .
ℎ 𝜕𝑌®𝑛+1
𝑘 ℎ
®¥𝑛
2𝑌 ®𝑛(𝑞)
𝑞𝑌
!
¤ ℎ ℎ
(55) 𝑧®𝑛 = 𝑌®𝑛 , ℎ𝑌®𝑛 , , ..., (60)
Compared to other numerical integration methods, within the 2! 𝑞!
implicit Euler method no intrinsic error estimation is possible.
There exist multistep algorithms that calculate an integration is stored, where 𝑌®𝑛 are the abundances at the current time,
¤ ¥
error by comparing the result of the integration after a full 𝑌®𝑛 , 𝑌®𝑛 , ..., 𝑌®𝑛(𝑞) are the time derivatives of the abundances,
time step with the result after two half steps. This how- and ℎ = 𝑡 𝑛+1 − 𝑡 𝑛 is the current step size. In order to obtain
(0)
ever increases the computational cost, and we therefore only the predictor step 𝑧®𝑛+1 , the Nordsieck vector is multiplied by
estimated the error based on a maximum change of the abun- a (𝑞 + 1) × (𝑞 + 1) Pascal triangle matrix defined by
dances 𝜖 Euler which is based on the current derivative. We
0 if 𝑖 < 𝑗

note that this procedure is similar to the time-step estimate 




of SkyNet (Lippuner & Roberts 2017). The approximate 𝑖𝑗
𝐴 (𝑞) = © 𝑖 ª with 𝑖, 𝑗 ∈ [0, 1, ..., 𝑞].
𝑖!
change within one time step is calculated by: 

 ­ ®= 𝑗!(𝑖− 𝑗 )! if 𝑖 ≥ 𝑗
 𝑗
« ¬
¤ 𝑌® (𝑡 + ℎ′ ) − 𝑌® (𝑡) (61)
𝑌® (𝑡) = (56) Therefore, the predictor step is given by
ℎ′
(0)
𝑌® (𝑡 + ℎ′ ) 𝑧®𝑛+1 = 𝑧®𝑛 · 𝐴, (62)
𝜖 Euler = max 1 − , (57)
𝑌® (𝑡) which is the Taylor series of 𝑌®𝑛 truncated at the order of 𝑞 in
therefore, we obtain matrix notation. To obtain an accurate solution for 𝑌®𝑛+1 the
predictor step is iteratively corrected due to
¤ (1 − 𝜖Euler )𝑌® (𝑡) − 𝑌® (𝑡) 𝑌® (𝑡)
𝑌® (𝑡) = ⇒ ℎ′ = 𝜖Euler max . (0)
𝑧®𝑛+1 = 𝑧®𝑛+1 ®
+ 𝑒®𝑛+1 · ℓ, (63)
ℎ′ ¤
𝑌® (𝑡)
(58) with the correction vector 𝑒®𝑛+1 . ℓ® is a 1 × (𝑞 + 1)-vector given
The default value in WinNet for 𝜖Euler is set to a maximum by
change of 10%. In order to avoid rapid changes of the time-
𝑞 𝑞  𝑞 
step, it is limited by the previous step size
 Ö 
∑︁
𝑗
Ö (𝑡 − 𝑡 𝑛+1 )/ℎ 𝑥
ℓ𝑗𝑥 = 1+ = 1+ .
! (𝑡 𝑛+1 − 𝑡 𝑛+1−𝑖 )/ℎ 𝜉𝑖
′ 𝑌® (𝑡) 𝑗=0 𝑖=1 𝑖=1
ℎ = min 𝐶 ℎ, 𝜖 Euler max (59) (64)
¤
𝑌® (𝑡) Here, we defined the vector 𝜉® storing the informa-
with the constant 𝐶 > 1. Furthermore, only species with tion of previous step sizes. The components of
abundances higher than a threshold abundance are taken into ℓ® = [ℓ0 (𝑞), ℓ1 (𝑞), ..., ℓ 𝑗 (𝑞), ..., ℓ𝑞 (𝑞)] are calculated as
account in the time step calculation (default in WinNet is ℓ0 (𝑞) = 1,
10−10 ). Additionally, the step size is restricted to a maximum 𝑞  
change of the density within one time step (default value
∑︁
ℓ1 (𝑞) = 𝜉𝑖−1 ,
5%) in order to get an adequate resolution for large density 𝑖=1
gradients. In case that nuclear heating is not enabled, the ℓ 𝑗 (𝑞) = ℓ 𝑗 (𝑞 − 1) + ℓ 𝑗 −1 (𝑞 − 1)/𝜉𝑞 ,
same restriction is applied to the temperature. ! −1
Ö 𝑞
3.2.2. Gear’s Method ℓ𝑞 (𝑞) = 𝜉𝑖 .
𝑖=1
In contrast to Euler’s method, Gear’s method (Gear 1971,
see also, e.g. Byrne & Hindmarsh 1975; Longland et al. The correction vector 𝑒®𝑛 is calculated using the same Newton-
2014; Martin 2017) includes terms of higher orders (see also Raphson scheme as for the solution 𝑌®𝑛+1 . To obtain the com-
 13

position of the next step, In addition to the automatic control of the step size, the order
  𝑞 can be selected automatically as well. For this, we allow
® (𝑚) = − 𝑌® (𝑚) − 𝑌® (0) + ℎ 𝑌®¤ (𝑚) − 𝑌®¤ (0) ,
ℎ    
1− 𝐽 Δ 𝑛+1 𝑛+1 𝑛+1 𝑛+1
only order changes of 𝑞 ±1. The error estimates for increasing
ℓ1 ℓ1
and decreasing order are calculated by:
(65)
(𝑚+1) (𝑚) ® (𝑚) 𝑞−1
𝑌®𝑛+1 = 𝑌®𝑛+1 +Δ (66) Î
𝜉𝑖 (𝑞)
𝑖=1 ℎ𝑞𝑌®𝑛+1
(0) ¤ (0) 𝐸 𝑛+1 (𝑞 − 1) = − (72)
is solved. Here, 𝑌®𝑛+1 and 𝑌®𝑛+1 are extracted from the first and ℓ1 (𝑞 − 1) 𝑞!
(0)
second entry of 𝑧®𝑛+1 . The index 𝑚 is the number of iterations, −𝜉𝑞+1 (𝑒 𝑛+1 − 𝑄 𝑛+1 𝑒 𝑛 )
𝐸 𝑛+1 (𝑞 + 1) =  , (73)
Δ (𝑚) is an iterative correction, and 𝐽 is the Jacobian matrix
 𝑞
Î 𝑡𝑛+1 −𝑡𝑛+1−𝑖
(𝑞 + 2)ℓ1 (𝑞 + 1) 1 + 𝑡𝑛 −𝑡𝑛+1−𝑖
(𝑚)
𝜕𝑌¤𝑖,𝑛+1 𝑖=2
𝐽𝑖 𝑗 = . (67)
(𝑚)
𝜕𝑌 𝑗,𝑛+1 where 𝑄 and 𝐶 are defined as
  𝑞+1
Identically to the implicit Euler integration, the Jacobian is 𝑄 𝑛+1 =
𝐶𝑛+1 ℎ 𝑛+1
(74)
represented by a sparse matrix, for which zero entries are 𝐶𝑛 ℎ𝑛
evaluated in every step. Calculating the Jacobian is one of Î𝑞
𝜉𝑖 " 𝑞
#
the most expensive steps when solving the ODE. Therefore, 𝑖=1
Ö 𝑡 𝑛+1 − 𝑡 𝑛+1−𝑖
some integration packages use so-called ’Jacobian caching’ 𝐶𝑛+1 = 1+ . (75)
(𝑞 + 1)! 𝑖=2
𝑡 𝑛 − 𝑡 𝑛+1−𝑖
to tackle the problem of recalculating the Jacobian multiple
times (e.g., VODE; Brown et al. 1989). We investigated a To obtain the most efficient way of calculating the solu-
similar technique, Broyden’s method (Broyden 1965), to ap- tion of the ODE, the step size in Eq. (70) is calculated
proximate the Jacobian instead of recalculating it in every for order 𝑞 − 1, 𝑞, and 𝑞 + 1, respectively. The or-
iteration. This, however, did not lead to a performance im- der is chosen as the one providing the largest time step,
provement due to rapid changes of the reaction rates and feed- ℎ′ = max(ℎ′ (𝑞 − 1), ℎ′ (𝑞), ℎ′ (𝑞 + 1)). Since the Nordsieck
back from the nuclear reactions on the temperature. There- vector depends on the step size (see Eq. 60), it must be rescaled
fore, more Newton-Raphson iterations were needed to obtain whenever the step size is changed:
convergence leading to an overall performance loss. After
the Newton-Raphson iteration has converged, the correction 𝑧®′𝑛+1 = diag(1, 𝜂, 𝜂2 , ..., 𝜂 𝑞 ) · 𝑧®𝑛+1 , (76)
vector
(0) where 𝜂 = ℎ′ /ℎ. Also, when the order decreases to 𝑞 − 1, the
𝑒®𝑛+1 = 𝑌®𝑛+1 − 𝑌®𝑛+1 (68)
Nordsieck vector has to be rescaled. Therefore, we define a
can be determined. To obtain a sophisticated guess of the correction
time step within a given tolerance, the error can be estimated ® ′ = 𝑑𝑖 𝑧®𝑞,𝑛+1 ,
Δ (77)
𝑖
by the truncation error
where, similar to Eq. (64), 𝑑® is implicitly defined as
" 𝑞   # −1
1 Ö 𝑡 𝑛+1 − 𝑡 𝑛+1−𝑖
𝐸 𝑛+1 (𝑞) = − 1+ 𝑒®𝑛+1 . (69) 𝑞
∑︁ 𝑞−2
Ö
ℓ1 𝑖=2
𝑡 𝑛 − 𝑡 𝑛+1−𝑖 𝑑 𝑗 𝑥 𝑗 = 𝑥2 (𝑥 + 𝜉𝑖 ) (78)
𝑗=0 𝑖=1
The next time step is computed within a certain allowed tol-
erance 𝜖 Gear by and its components are given by:
  1/𝑞+1
ℎ′ = ℎ𝐾
𝜖 Gear
, (70) 𝑑0 (𝑞) = 𝑑1 (𝑞) = 0
max 𝐸¯ 𝑛+1 (𝑞) 𝑞−2
Ö
where 𝐾 is a conservative factor usually chosen in the inter- 𝑑2 (𝑞) = 𝜉𝑖 ,
val 𝐾 ∈ [0.1, 0.4]. As for the calculation of the step size in 𝑖=1

Eq. (59), only abundances above a certain threshold should 𝑑 𝑗 (𝑞) = 𝜉𝑞−2 𝑑 𝑗 (𝑞 − 1) + 𝑑 𝑗 −1 (𝑞 − 1),
contribute to the calculation of the new time step. Therefore, 𝑞−2
∑︁
the truncation error is rescaled in order to prevent an over- 𝑑 𝑞−1 (𝑞) = 𝜉𝑖 ,
weighting of the change of very small abundances, smaller 𝑖=1
than a threshold 𝑌limit (default in WinNet: 10−10 ), 𝑑 𝑞 (𝑞) = 1.

Due to the implementation of higher orders, within Gear’s


 𝐸 𝑖,𝑛+1 /𝑌𝑖

 if 𝑌𝑖 > 𝑌limit
𝐸¯ 𝑖,𝑛+1 = . (71) method, one is able to apply larger step sizes compared to
 𝐸 𝑖,𝑛+1 /𝑌limit
 if 𝑌𝑖 ≤ 𝑌limit

14

the implicit Euler scheme, reducing the amount of iterations Tinit

NSE
drastically without losing accuracy. However, for most of the
calculations, more Newton-Raphson iterations are necessary, TNSE, h
resulting in similar or even higher computational costs. The
TNSE, c

Temperature
difference between the implicit Euler and Gears method is

intermediate
discussed in more detail in Appendix A. Tra
jec
4. REACTION NETWORK INPUTS tor
y
4.1. Lagrangian tracer particles
Expansion T
The nuclear reaction network equations (Eq. 20) contain cold

cold
a dependency on the temperature and density of the envi-
ronment. To get an initial composition from NSE, addition-
Time
ally the electron fraction is necessary. These quantities have
therefore to be recorded from a simulation of an astrophysical
scenario. This is often done in terms of Lagrangian tracer Figure 4. Sketch of different temperature regimes included in Win-
particles within the hydrodynamic simulation. These parti- Net.
cles (also called trajectories or tracer) are passively advected
within the (M)HD simulation, tracing all relevant quantities old 𝑇NSE , the nuclear reaction network is solved for all nuclear
such as the time, temperature, density, electron fraction, and reactions. The transition temperature between these regimes
neutrino properties. WinNet is a so-called single-zone code, can be chosen individually, depending on whether the tran-
i.e., tracer particles cannot interact among one another. This sition occurs from hot to intermediate temperatures (𝑇NSE,c )
assumption is valid if the nuclear burning timescales are much or from intermediate to hot temperatures (𝑇NSE,h , see Fig. 4).
faster than other timescales changing the abundances (e.g., The exact temperatures of the transitions depend on the en-
diffusion). Therefore, for the majority of explosive environ- vironment (e.g., Khokhlov 1991). The reason for having
ments we can use a single-zone reaction network; however, two transition temperatures is mainly motivated when using
for some cases such as hydrostatic oxygen burning, it has to a feedback of the nuclear energy on the temperature (Sec-
be taken with care (Hix & Thielemann 1999). There have tion 2.4). In this case, a slight inconsistency at the interface
been several studies on the uncertainties of a tracer particle between the hot and intermediate regime (see Section 2.2)
method. The necessary amount of tracer particles to achieve may cause fluctuations in the temperature that can lead to an
convergence has been studied, e.g., in Seitenzahl et al. (2010) infinitesimal time step when using only one transition tem-
and Nishimura et al. (2015). Also, the initial placement of perature.
the tracer particles can have an impact on the convergence of When the temperature drops below 𝑇 = 10−2 GK, all reac-
the result (Bovard & Rezzolla 2017). A detailed compari- tion rates are frozen to the lower validity limit of the JINA
son between setting tracers in contrast to calculating the nu- Reaclib reactions (brown region in Fig. 4, Cyburt et al. 2010).
cleosynthesis inside the hydrodynamical simulation has been Often, the Lagrangian tracer particle finishes before the nu-
presented in the context of CC-SNe by Harris et al. (2017) and cleosynthesis is completed and an extrapolation of the ther-
by Navó et al. (2023). Additionally, Sieverding et al. (2023) modynamic conditions is required (dotted line in Fig. 4). The
studied the impact of obtaining tracers in a post-processing details of these assumptions can have an impact on the final
step after the calculation of a hydrodynamic model from sim- yields (see also Harris et al. 2017) and should be chosen ac-
ulation snapshots. cording to the environment, e.g., a homologous expansion for
CC-SNe or a free expansion for the dynamical ejecta of an
4.1.1. Temperature and density regimes Neutron star merger (NSM).
During its evolution, a tracer particle can undergo differ-
ent temperature regimes, and therefore different approaches 4.2. Reaction rates
are required to obtain the composition within the given time Although all nuclei are connected to each other by nuclear
step. In WinNet, there are distinctions between three temper- reactions, in practice most of the reactions are negligible.
ature regimes, the regime of NSE, the intermediate tempera- The most important reactions for astrophysical environments
ture regime, and the cold temperature regime, schematically are given by reactions that involve nucleons or 𝛼-particles,
shown in Fig. 4. In the regime of NSE, the network equa- decays, neutrino reactions, electron- and positron-captures,
tions are only solved for weak reactions. Instead of calculating or fission reactions (Fig. 6). There exist many formats of the
also strong reactions, an equilibrium is assumed (Section 2.3). reaction rates. WinNet is built around the Reaclib reaction
When the conditions are below a certain temperature thresh- rate library, and this library usually contains the majority of
 15

Reading all reaction Z+1,N Z+2,N+1

rates

(γ, p)
(n, (p, γ)
Z+2,N+2
Z+1,N-1

(α , α)
(n n)

(α , α )

)
Initialize rate array with

n( )
γ
,γ
, p ,(β
(fi

(p )
Reaclib rates Z+1,N+2

),( −
ss

,
α)

io

β ν)
+ν
(p, p)

)
Add/merge
(n, γ) (α, (n, γ)
Z,N-1 Z,N Z,N+1
alpha-decay rates (γ, n) , α) (γ, n)

(n , n)
(p p)

Protons

, p ,(β
(p
(α,

(fi )
),( −
(α α)

ss
Z-1,N-2

(γ, p) , n)

β ν)
Add/merge

+ν
T < Texp

(α , α )

io
(n) ,
β-delayed neutron

n)
(γ
,γ
emission rates

(p, γ)
Z-2,N-2 Z-1,N+1
Increasing priority

Add/merge
tabulated rates Z-2,N-1 Z-1,N

Add/merge
+ -
Neutrons
theoretical β , β ,
ec, and pc rates

Figure 6. Sketch of the most important nuclear reactions (Reichert

Add 2021).
neutrino reactions
with the highest priority as in Fig. 5 or by deleting unwanted
Add rates from high-priority formats. The modular structure of
fission reactions WinNet allows for an easy implementation of other popu-
lar reaction rate formats. In the following, we give a short
Add/merge overview of the current supported file formats.
detailed balance
reactions
4.2.1. Reaclib file format
Most of the nuclear reaction rates are given in form of seven
Figure 5. Sketch of the rate replacement procedure. Reaction rates fit parameters, 𝑎 𝑖 , the so-called Reaclib14 format (Cyburt et al.
in different formats have different priorities when creating a list with 2010). The reaction rate is calculated according to:
all reactions within WinNet. The priority of the rates increases from
" 5
#
the top to the bottom of the plot. At a certain threshold temperature ∑︁ 2𝑖−5
𝑇exp , theoretical 𝛽+ , 𝛽 − , ec, and pc rates get replaced again as 𝑅 = exp 𝑎 0 + 𝑎 𝑖 𝑇9 3 + 𝑎 6 ln𝑇9 . (79)
they are only valid above certain temperatures (see text). With the 𝑖=1
exception of Reaclib rates, all other rates are only optionally used.
Depending on the reaction, 𝑅 can be either 𝜆, 𝑁A ⟨𝜎⟩𝑖, 𝑗 , or
𝑁A2 ⟨𝑖 𝑗 𝑘⟩ (see Eq. 20). Reverse reactions have additionally
reactions (Cyburt et al. 2010). However, other formats are to be multiplied by the partition functions (see the pre-factor
also supported, e.g., tabulated reaction rates from the TALYS in Eq. 21) that are also provided within the Reaclib database
code (Koning et al. 2019). Rates given in different formats are in a separate file (with the data originating from Rauscher
either added or merged into the list of all rates within WinNet. & Thielemann 2000 and Cyburt et al. 2010 for more proton-
In this case, the different formats have different priorities, rich nuclei). For higher temperatures (𝑇 > 10 GK) partition
starting with the Reaclib reactions with the lowest priority. functions from Rauscher (2003) can be used. The fits of the
If, in addition, this rate is also included in the theoretical 𝛽+ , reaction rates are valid between 10−2 GK ≤ 𝑇 ≤ 102 GK. For
𝛽 − , ec, and pc rates, it is replaced once again. The priority of lower temperatures, within WinNet, the rates are kept con-
the individual rates is shown in Fig. 5. stant. At higher temperatures, usually NSE is assumed that
We note that WinNet does not perform any evaluation on only depends on the binding energies and partition functions.
the reliability of a rate. If a rate is contained multiple times Each reaction belongs to a specific chapter in the Reaclib
in different formats13, it is the user’s responsibility to choose tables as given in Table 1. The Reaclib chapters correspond to
the desired rate by either fully automatically using the one

14 See https://reaclib.jinaweb.org/index.php and Rauscher & Thielemann

13 Not to confuse with a rate being contained multiple times in the same format, (2000); Thielemann (1980) for more details about the format and for re-
which can happen due to, e.g., resonances in the rate. cent reaction rates
16

Chapter No. Reactants No. Products

100 101

Proton number
1 1 1
100

Q (MeV)
80
2 1 2
0
3 1 3
4 2 1
60 100
5 2 2 40 101

-decay half-lives (s)

6 2 3 100 108

Proton number
7 2 4
8 3 1
80 104
9 3 2 60 100
10 4 2
40 10 4
11 1 4
100

Proton number
Table 1. Amount of reactants and products for different Reaclib
chapters. 80
Stable
60 Experiment
Parameterization
different one-, two-, and three-body terms in Eq. (20), where No -decays
40
each of these terms in the summation can include different 50 75 100 125 150 175 200
numbers of reaction products. Another Reaclib format in- Neutron number
cludes Chapter 8 and 9 together and does not include Chapter
10 and 11. WinNet supports and automatically detects both
options. We note that the Reaclib reactions also contain two Figure 7. Upper panel: Q-value for 𝛼-decay using the masses
provided with the Jina Reaclib. Second panel: 𝛼-decay half-lives
isomers of 26 Al. WinNet can take these isomers into account
in seconds. Whenever experimental 𝛼-decay half-lives are avail-
when adding their properties into the winvn and in the list able, we plot these instead of the parameterized ones. Nuclei that
of considered nuclei. They will then be treated like all other have half-lives of 𝑇1/2 ≳ 1012 yr are assumed to not 𝛼-decay. Bot-
nuclei. tom panel: distinction between parameterized and experimentally
known 𝛼-decay half-lives included in the Jina Reaclib. Stable nu-
4.2.2. Parametric 𝛼-decays clei are shown as black squares, experimentally available 𝛼-decays
The Reaclib reaction rate database contains only experi- within the Jina Reaclib are indicated as dark-gray rectangles. Magic
mental 𝛼-decays. To make the 𝛼-decays more complete, Win- numbers of 50, 82, and 126 are shown as dashed lines. All shown
rates are publicly available along with WinNet.
Net is able to calculate additional 𝛼-decays with the Viola-
Seaborg formula (e.g., Viola & Seaborg 1966; Sobiczewski
et al. 1989; Brown 1992; Sahu & Bhoi 2016). We provide rate words, they are more relevant for regions with high 𝑄 𝛼 (see,
tables of 𝛼-decays using the parameterization of Dong & Ren upper and middle panels of Fig. 7). It has been pointed out
(2005). For 𝑍 > 84 and 𝑁 > 126, they fitted experimentally that this fit is only valid for their fitting regions; other regions
determined 𝛼-decays with need a separate fit. To also obtain a valid fit to the other
regions, we use the masses and experimental 𝛼-decay half-
log10 𝑇𝛼 = (𝑎𝑍 + 𝑏)𝑄 −0.5
𝛼 + (𝑐𝑍 + 𝑑) + ℎ𝑙𝑜𝑔 , (80) lives of the Reaclib. We therefore use the above parameters
only for nuclei with 𝑍 ≥ 82 and 𝑁 ≥ 126, while we use the
where 𝑍 is the proton number of the decaying nucleus, 𝑄 𝛼 parameters of Table 2 for the other regions.
is the Q-value of the decay, and the parameters 𝑎 = 1.64062, This fit over these four individual regions of the nuclear
𝑏 = −8.54399, 𝑐 = −0.19430, and 𝑑 = −33.9054 were chart that correspond to the regions between magic num-
derived through least-squares fitting. Additionally, the so- bers is in a much better agreement to the experimental half-
called hindrance factor ℎlog was fitted: lives (Fig. 8). Still, some deviations of around 1–2 magni-
tudes are present around the magic numbers. When compar-
0, Z even, N even

ing all available experimental 𝛼-decays with the calculated




ones, we obtain a standard deviation of 𝜎Zeven,Neven = 0.38,

 0.8937,

 Z even, N odd
ℎlog = . (81) 𝜎Zodd,Neven = 1.61, 𝜎Zeven,Nodd = 0.93, and 𝜎Zodd,Nodd = 0.82.
 0.5720,
 Z odd, N even


 The large standard deviation of 𝜎Zodd,Neven is driven by the
decay of 153 Lu whose half-life differs by more than 16 mag

 0.9380,
 Z odd, N odd
(3.9 × 1016 s versus an experimental value of ∼ 1.3 s). Note
An obvious consequence of this parameterization is that 𝛼- that 153 Lu has a magic neutron number of 82; nevertheless,
decays happen on shorter timescales if 𝑄 𝛼 is large or, in other
 17

Neutron number the difference between the Viola-Seaborg formula and the
60 80 100 120 140 160 experimental value is quite remarkable and indeed possibly
a result of an outdated rate in the Jina Reaclib that uses the
2 experimental data last evaluated in 2017. The latest experi-
mental data from 2019 indicates that this nucleus is entirely
0
decaying by an ec/𝛽+ -decay15 which would agree with the
)
log10 T1/2, exp/T1/2, calc

2 large half-life obtained with our fitted formula. When remov-

ing this nucleus from the calculation of the standard deviation,
it reduces to 𝜎Zodd,Neven = 0.64. We therefore have excluded
2 it from our least-squares fit.
(

On a technical level, within WinNet one can decide if the

0 𝛼-decay rates should only supplement the Reaclib rates or also
replace them. The latter may become interesting in the future
2 N odd, Z odd N even, Z odd
N odd, Z even N even, Z even in case other theoretical 𝛼-decays will be added to the Reaclib.
In addition, one can adjust between which proton numbers 𝛼-
50 60 70 80 90 100 110 decays are added. Within WinNet we provide a file with the
Proton number 𝛼-decay rates using the parameterization presented here. For
the fit as well as the rates, we used the masses of the Jina
Figure 8. Ratio of calculated and experimental 𝛼-decay half-lifes. Reaclib as an input.
The upper panel shows the ratio versus neutron number, the lower 4.2.3. Tabulated rates
panel versus proton number. The different colors indicate the differ-
ent types of nuclei as indicated in the legend. Another possible format is given in form of a tabulation.
This format is common for nuclear reaction codes such as
TALYS (Koning et al. 2019). Every rate is tabulated on 30
temperature grid points from 10−4 to 10 GK and, identical
to the Reaclib format, assigned a certain chapter as given
a b c d in Table 1. Reaction rates that are given in tabulated form
𝑍 > 82 will replace the respective reaction rates in Reaclib format.
1.64062 -8.54399 -0.19430 -33.9054 Reverse reactions can be given in tabulated form or calculated
𝑁 > 126
with the theory of detailed balance within WinNet. These
𝑍 > 82
1.71183 -7.50481 -0.25315 -30.7028 calculations will replace all reverse rates that are given in the
82 < 𝑁 ≤ 126
reaction rate library.
50 < 𝑍 ≤ 82
1.70875 -7.52265 -0.25153 -30.8245
82 < 𝑁 ≤ 126 4.2.4. Neutrino reactions
50 < 𝑍 ≤ 82 Neutrino reactions are tabulated versus the neutrino tem-
1.71371 -7.34226 -0.24978 -30.6826
50 < 𝑁 ≤ 82 peratures from 2.8 to 10 MeV on seven grid points. These
h1 h2 h3 h4 reaction rates enter the nuclear reaction network as an addi-
𝑍 > 82 tional term in the form of
0 0.8937 0.5720 0.9380
𝑁 > 126 D𝑌 (𝑡)
= ⟨𝜎⟩(𝑡) 𝐹𝜈 (𝑡) 𝑌 (𝑡), (82)
𝑍 > 82 D𝑡
0 0.0476 0.1214 0.3933
82 < 𝑁 ≤ 126 with the average neutrino cross section integrated over the nor-
50 < 𝑍 ≤ 82 malized neutrino spectrum ⟨𝜎⟩(𝑡) that depends on the neu-

0 0.2140 0.0600 0.4999
82 < 𝑁 ≤ 126 trino temperature 𝑇𝜈 (𝑡). Furthermore, 𝐹𝜈 = 𝐿 𝜈 / 4𝜋𝑟 2 ⟨𝐸 𝜈 ⟩
50 < 𝑍 ≤ 82 is the neutrino number flux.
0 -0.1242 1.1799 0.7166
50 < 𝑁 ≤ 82 WinNet includes a tabulation where the neutrino reac-
tions on nucleons have been calculated as described in, e.g.,
Table 2. Fitted parameters for Eq. (80). For 𝑍 > 82, 𝑁 > 126 we Burrows et al. (2006) with the weak magnetism and recoil
use the parameterization of Dong & Ren (2005). The lower part of corrections as in Horowitz (2002). Within WinNet we pro-
the table shows the hindrance factors and h1 indicates Z even and N
vide the rate table as well as a python script to calculate
even, h2 Z even and N odd, h3 Z odd and N even, h4 Z odd and N
odd.
15 https://www-nds.iaea.org/exfor/servlet/E4sGetIntSection?SectID=
14658963&req=2130
18

it. In principle the full neutrino energy distribution could 104

80 Charged-current
of
reactions
be taken from the hydrodynamic simulation and an appro- e
priate neutrino temperature can be calculated based on this. 60
In WinNet, the average neutrino energy ⟨𝐸 𝜈 ⟩ is used inter-
changeably with the neutrino temperature 𝑇𝜈 by assuming a 40 103
Fermi-Dirac distribution of the neutrino energies and a zero
chemical potential of the neutrinos. For this case, 20
F3 (0) 7𝜋 4 0 102
⟨𝐸 𝜈 ⟩ = 𝑇𝜈 = 𝑇𝜈 ≈ 3.1513 𝑇𝜈 (83) 80 Charged-current
of
reactions
F2 (0) 180 𝜁 (3) e

(10 42cm2)
holds. Here 𝜁 is Riemann’s zeta function, and F𝑛 are the 60
Fermi integrals defined as 40 101
∫ ∞
1 𝑥𝑛

Proton number
F𝑛 (0) =
Γ(𝑛) 0 exp (𝑥) + 1
d𝑥, (84) 20
with the gamma function Γ(𝑛) = (𝑛 + 1)!. We note that 0 100
current CC-SNe simulations hint toward slight deviations of
80 Neutral-current
of
reactions
i
the Fermi-Dirac distribution. Such a deviation can have an 60
impact on the energy integrated neutrino cross sections that 10 1
we do not take into account with the provided tabulation (e.g., 40
Tamborra et al. 2012; Mirizzi et al. 2016; Sieverding et al.
2019). 20
For neutrino reactions with heavier nuclei, WinNet is able 0 10 2
to include neutrino interactions that are provided in a sepa- 80 Neutral-current reactions 4
of i
rate file. This file is taken from Sieverding et al. (2018) and
60 3
includes charged-current as well as neutral-current reactions.

<n>
All of these reactions contain different reaction channels al- 40 2
lowing for the ejection of light particles such as neutrons,
protons, and an alpha-particle. An overview of these cross 20 1
sections is illustrated in Fig. 9, where we show the cross
sections summed over all reaction channels and the average 0 0
0 25 50 75 100 125 150
amount of ejected neutrons for neutral-current reactions at Neutron number
𝑇𝜈 = 5 MeV.
Including neutrinos in the calculation requires additional
information in form of either a tabulation or a parameteri- Figure 9. Energy averaged neutrino cross sections from the table
zation of these neutrino properties. In the case of charged- of Sieverding et al. (2018) at 𝑇𝜈 = 5 MeV. Shown are the summed
cross sections of all reaction channels. The individual panels show
current reactions, only the properties of electron neutrinos and
charged-current reactions of electron neutrinos 𝜈𝑒 , charged-current
antineutrinos have to be provided. Neutral-current reactions reactions of electron antineutrinos 𝜈¯𝑒 , neutral-current reactions of
need additional properties of muon and tau (anti)neutrinos. any neutrino flavor 𝜈𝑖 , and the average amount of neutrons for
Within WinNet it is assumed that the (anti)neutrino energies neutral-current reactions of any neutrino flavor 𝜈𝑖 . Note that the
(or temperatures) for muon and tau neutrinos are identical properties of neutral-current reactions of any antineutrino flavor 𝜈¯𝑖
(𝐸 𝜈𝜇 = 𝐸 𝜈𝜏 ), and they are thus included as species 𝑥, where are nearly identical to the lower two panels.
𝐸 𝜈𝑥 has to be provided. Furthermore, the summed luminosi-
ties have to be provided (𝐿 𝜈𝑥 = 𝐿 𝜈𝜇 + 𝐿 𝜈𝜏 ) for neutrinos and listed on a temperature and electron density grid (e.g., Fuller
antineutrinos. Treating muon and tau (anti)neutrinos effec- et al. 1985; Oda et al. 1994; Langanke & Martı́nez-Pinedo
tively together as described above may be sufficient, as current 2001; Pruet & Fuller 2003; Suzuki et al. 2016). A direct tab-
CC-SNe simulations do not really distinguish between these ulation of the rates, however, can lead to large interpolation
neutrino flavors, and little has been done in this direction so errors (Fuller et al. 1985). Therefore, the rates are not tabu-
far (however, see Bollig et al. 2017). lated directly, and instead, effective log 𝑓 𝑡 eff is stored. This
4.2.5. Theoretical weak rates can be converted to the actual rate via (see, e.g., Langanke &
Martı́nez-Pinedo 2001):
Theoretical models, e.g., shell-model calculations, are used
𝐼
to obtain weak rates for stellar conditions. These rates are 𝜆 = ln 2 . (85)
𝑓 𝑡eff
 19

Here, 𝐼 is the phase space integral for ground-state to ground-

state transitions 40
∫ ∞
𝜔2 (𝑞 + 𝜔) 2 𝑆(𝜔)𝑑𝜔, 30

Proton number
𝐼= (86)
𝜔0 =max(𝑞,1)

with 𝑞 = (𝑚 𝑖 − 𝑚 𝑓 )/𝑚 𝑒 the Q-value in units of the electron 20 FFN

mass. O
1 LMP
𝑆(𝜔) = , (87) 10
exp 𝜔𝑚𝑒 𝑐 2 −𝜇𝑒
+ 1 PF
STN
𝑘B 𝑇

with the electron chemical potential 𝜇𝑒 . 0

We note that these theoretical reaction rates usually ne- 0 10 20 30 40 50 60 70
glect atomic electron-capture, which becomes increasingly Neutron number
important for lower temperatures, e.g., for 56 Ni. Therefore,
WinNet contains the possibility of replacing all theoretical
Figure 10. Compiled file of theoretical 𝛽 − -, 𝛽+ -decays, and
decays, electron- and positron-captures at low temperatures positron- and electron-capture rates originating from different
with the experimental decays provided in the Reaclib. sources. The sources are FFN (Fuller et al. 1985), O (Oda et al.
WinNet supports an individual grid for each reaction for 1994), LMP (Langanke & Martı́nez-Pinedo 2001), PF (Pruet &
the tabulation of theoretical 𝛽 − -, 𝛽+ -decays, positron- and Fuller 2003), and STN (Suzuki et al. 2016). Stable nuclei are indi-
electron-capture rates. This is necessary, as different avail- cated as black boxes.
able tabulations were calculated on different temperature and
log 𝜌𝑌𝑒 grids. We provide a table that was compiled out of format. Additionally, there exist user-defined parameters to
various sources covering different regions of the nuclear chart allow for a controlled replacement of rates. With them, one
(Fig. 10)16. Note that WinNet also uses electron-capture rates can specify if, e.g., experimentally measured decays should
on protons as well as positron-captures on neutrons from this also be replaced.
table.
Since the reaction rates are tabulated with a dependence on 4.2.7. Fission reactions and fragments
the electron density, in principle the derivative of the reaction There are various fission modes, of which WinNet in-
rate with respect to the abundances should be nonzero and cludes three: spontaneous fission, neutron-induced fission,
there should be a term representing this in the Jacobian of the and beta-delayed fission. In all of these cases, in addition to
system (Eq. 52 and Eq. 67). Within WinNet we ignore this the probability to undergo fission, the resulting fission frag-
dependence and assume a zero derivative of these reaction ment distribution is of importance as well. Investigations
rates. for fission barrier heights utilized in astrophysics have been
4.2.6. 𝛽-delayed neutron emission performed from 1980 until today (Howard & Möller 1980;
Myers & Świaţecki 1999; Mamdouh et al. 2001; Goriely
The Reaclib file format only allows 𝛽-delayed neutron emis- et al. 2009; Giuliani et al. 2018a,b; Vassh et al. 2019; Giu-
sions up to three neutrons (Reaclib chapter 11; see Table 1). liani et al. 2020). Neutron-induced cross section predictions
In practice, decays that emit only up to two neutrons are in- (or also beta-delayed fission) for astrophysical applications
cluded. The probability of all other decay channels in the were treated (by, e.g., Panov et al. 2005; Martinez-Pinedo
Reaclib format is added to the decay channel with three prod- et al. 2007; Panov et al. 2010; Erler et al. 2012; Giuliani et al.
ucts. Especially when matter far from the valley of stability 2018a. Extended compilations have been provided and can
on the neutron-rich side is synthesized, 𝛽-delayed neutron be found in several databases17.
emission of more than two neutrons can occur (e.g., Marketin In the present paper, we provide only a limited set of fis-
et al. 2016; Möller et al. 2019). Therefore, WinNet supports sion inputs, which are available within the WinNet package
a file format containing the half-lives of the nuclei and the and are stored in a separate file. The format is similar to the
different channel probabilities up to the 𝛽-delayed emission Reaclib file format, but only the name of the parent nucleus
of 10 neutrons. Optionally, average emitted neutrino energies is stored. WinNet includes the rates of Panov et al. (2005)
can be provided in this file (to account for the energy loss for 𝛽-delayed fission, and Panov et al. (2010) for neutron-
when self-heating is enabled; see Section 2.4). Duplicates in induced fission. Reaction rates for spontaneous fission have
Reaclib format will be replaced by the reaction rates in this

17 https://nucastro.org, https://www.jinaweb.org/science-research/scientific-
16 See https://groups.nscl.msu.edu/charge exchange/weakrates.html resources/data and https://www-nds.iaea.org, including TALYS results.
20
130
1023 ing neutron star (Schatz et al. 2002), complete Si burning
125 within a CCSN (with a simple parametric model as described
120 1016 in Nadyozhin & Deputovich 2002; Woosley et al. 2002) the
Proton number

115
109 neutrino-driven wind within a CCSN (Bliss et al. 2018), the

T1/2, sf (s)
110 detonation phase of a type Ia supernova (with a parametric
105 102 model as in Meakin et al. 2009), a main 𝑠-process (Cescutti
100
10 5 et al. 2018; Cescutti 2022), a weak 𝑠-process (Hirschi et al.
95 2004; Nishimura et al. 2017a; Pignatari & Hirschi 2022),
90 10 12
hydrostatic hydrogen burning, carbon-oxygen burning, and a
140 150 160 170 180 190 200
Neutron number simple 𝑖-process model (as described in Dardelet et al. 2015).
All of these conditions are examples in WinNet and should
guide the user on how to use the code. It is noteworthy that the
Figure 11. Half-lives of spontaneous fission in the nuclear chart
trajectories represent typical conditions in the scenarios and
(c.f., Khuyagbaatar 2020). Colored dots indicate experimentally
measured half-lives taken from the ENDFS database. may be used for sensitivity studies, but they do not necessarily
reflect the total yields that can be obtained when calculating
been calculated with the semi-empirical formula of Khuyag- often thousands of trajectories of the individual scenarios.
baatar (2020), using the fission barriers provided in Möller Furthermore, a different nuclear physics input is used within
et al. (2015). These half-lives together with experimentally the example cases, and we do not aim to exactly reproduce
measured ones are shown in Fig. 11. While in Khuyagbaatar the abundances that have been obtained within the original
(2020) spontaneous fission half-lives were fitted to nuclei publications. All example cases are very diverse in their in-
with even neutron and proton numbers only, we use the same volved conditions and together they cover a large range of the
equation for all nuclei. nuclear chart. In the following sections we present only a
The products (or fission fragments) are described by a fis- subset of the aforementioned examples.
sion fragment distribution in a probabilistic way. They can
either be described by an analytic formula (Kodama & Taka- 5.1.1. Big Bang nucleosynthesis
hashi 1975; Panov et al. 2001) or more complicated models The synthesis of elements during the first minutes after the
(e.g., Kelic et al. 2009; Goriely et al. 2009; Mumpower et al. origin of our Universe can be calculated with a relatively small
2020). Within WinNet we include the fragment distribution network. Following Winteler (2012), we create a trajectory
of Kodama & Takahashi (1975), Panov et al. (2001), and for a flat, isotropic, and homogeneous Universe to describe
Mumpower et al. (2020). As pointed out in Mumpower et al. the conditions during the big bang (see also Vonlanthen et al.
(2020), the distribution should only be used for 𝛽-delayed 2009). Furthermore, we assume a freeze-out of weak reac-
and neutron-induced fission. Therefore, WinNet contains tions at 𝑇 = 0.8 MeV. An important quantity is the initial
these fragment distributions in combination with the ones of photon-to-baryon ratio, which was measured by the Planck
Kodama & Takahashi (1975) for spontaneous fission. Satellite (5.96 × 10−10 ≤ 𝜂 ≤ 6.22 × 10−10 , Planck Collab-
5. REACTION NETWORK APPLICATIONS
oration et al. 2016). By creating one trajectory for each
baryon to photon ratio we are able to connect the big bang
5.1. Example cases nucleosynthesis with measurements of abundances in stars
In the following, we discuss several example cases calcu- and therefore probe the conditions of the big bang. For
lated with WinNet that are available together with the code. deuterium, the primordial abundance was determined to be
These examples involve conditions of a variety of scenarios, 𝑌 (D)/𝑌 (H) = (2.527 ± 0.03) × 10−5 (Cooke et al. 2018,
namely the Big Bang (as described in Winteler 2012), the dy- orange band in Fig. 12). For deuterium there is a slight
namic ejecta of an NSM (from Korobkin et al. 2012; Rosswog discrepancy with respect to the photon-to-baryon ratio deter-
et al. 2013; Piran et al. 2013; Bovard et al. 2017), the neutrino- mined by the Planck Satellite and observed deuterium abun-
driven wind of an NSM (Perego et al. 2014; Martin et al. dances. As the deuterium abundance is very sensitive to
2015), the viscous disc ejecta of an NSM (Wu et al. 2016; Lip- the d(p,𝛾)3 He reaction rate, this discrepancy may vanish in
puner et al. 2017), and the dynamic ejecta of a black hole neu- the future with new experimentally determined reaction rates
tron star merger (Korobkin et al. 2012; Rosswog et al. 2013; (Mossa et al. 2020; Moscoso et al. 2021). Here, we used
Piran et al. 2013). Additionally, we provide various condi- the rate of Descouvemont et al. (2004) that is included in the
tions within MR-SNe (Winteler et al. 2012; Obergaulinger JINA Reaclib. Furthermore, observations of 𝑌 (D)/𝑌 (H) are
& Aloy 2017; Aloy & Obergaulinger 2021; Obergaulinger & also differing (e.g., Romano et al. 2003). The observed value
Aloy 2021; Reichert et al. 2021, 2023), classical novae (José of 𝑌 ( 3 He)/𝑌 (H) = (1.1 ± 0.2) × 10−5 (Bania et al. 2002) is
& Hernanz 1998; José 2022), the X-ray burst of an accret- in perfect agreement with the estimated value. Additionally,
 21

He 4 WinNet
10 1 10 1
Cescutti et al. 2018
Initial mass fractions
10 3
Abundance relative to hydrogen

10 3 D
He 3

Mass fractions
10 5

10 5
10 7

10 Planck Satellite
7
10 9

Li 7 + Be 7
10 9 10 11
0 25 50 75 100 125 150 175 200
Mass number

10 11
10 12 10 11 10 10 10 9 10 8 10 7 Figure 13. Initial and final mass fractions of a main 𝑠-process. The
trajectory as well as the final mass fractions are taken from Cescutti
et al. (2018) accessed via Cescutti (2022).

Figure 12. Final abundances relative to hydrogen as a function of

the photon-to-baryon ratio 𝜂. Horizontal bands show measurements where the shock temperature 𝑇𝑆 can be defined as in, e.g.,
of the respective isotope. Nadyozhin & Deputovich (2002); Woosley et al. (2002)
1/4 −3/4
𝑇𝑆 = 2.4 𝐸 51 𝑅0 GK, (91)
the value of 𝑌 ( 4 He) = 1/4 × (0.2561 ± 0.0108)/𝑌 (H) (Aver
et al. 2010) is in agreement with our calculation (blue band with the explosion energy 𝐸 51 in 1051 erg, and an initial ra-
in Fig.12). The observed 7 Li abundance (𝑌 ( 7 Li)/𝑌 (H) = dius 𝑅0 in 108 cm. The shock density is given by the jump
+0.68 , Ryan et al. 2000) is in clear discrepancy with the
1.23−0.32 condition (𝜌 𝑆 = 7𝜌0 ). For an initial (pre-shock) density of
calculated value. This well-known problem is referred to in 𝜌0 = 106 g cm−3 , an initial radius of 𝑅0 = 2 × 108 , and an
literature as the lithium problem (see, e.g., Fields 2011; Fields explosion energy of 1052 erg we obtain:
& Olive 2022, for reviews).
𝑇 (𝑡) = 2.4 (0.2) −3/4 𝑒 −𝑡/(3𝜏 ) (92)
5.1.2. Main 𝑠-process 6 −𝑡/𝜏
𝜌(𝑡) = 7 × 10 𝑒 . (93)
We added a trajectory of a main 𝑠-process to the example
cases. This trajectory was used for a Monte Carlo sensitivity When we further assume an electron fraction of 𝑌𝑒 = 0.498
study in Cescutti et al. (2018) and can be accessed via Cescutti as typical in the Si shell, we obtain final abundances that are
(2022). The trajectory was extracted from the 13 C pocket after located around 56 Fe (Fig. 14).
the sixth thermal pulse of a solar metallicity, 3 M⊙ mass AGB
star (for more details, see the original publication). The final 5.1.4. 𝜈p-process
mass fractions agree well with those of Cescutti et al. (2018, Neutrinos can be crucial to synthesize proton-rich isotopes.
see Fig. 13) given the fact that we do not attempt to use the If the neutrino flux is strong enough, this can lead to a 𝜈p-
exact same nuclear input. process. The conditions for this are, for example fulfilled
in the MR-SNe model 35OC-RO of Obergaulinger & Aloy
5.1.3. Complete Silicon burning
(2017) and Reichert et al. (2021). The nucleosynthetic flow
The complete Si burning can be described by analytical with and without neutrinos is shown in Fig. 15.
models. For this, we assume that the time scale behaves
according to the freefall time scale(e.g., Arnett 1996): 5.1.5. The weak 𝑟-process
446 The weak 𝑟-process, i.e., a synthetization of elements up
𝜏≈ √ . (88) to the second 𝑟-process peak (𝐴 ∼ 130) can occur in moder-
𝜌
ately neutron-enriched environments. These conditions can
The density 𝜌 and the temperature 𝑇 are assumed to follow: be found in a variety of astrophysical host scenarios. Here,
𝑇 (𝑡) = 𝑇𝑆 𝑒 −𝑡/(3𝜏 ) (89) we show an exemplary trajectory from an MR-SNe (Ober-
−𝑡/𝜏
gaulinger & Aloy 2017; Reichert et al. 2021), the neutrino-
𝜌(𝑡) = 𝜌 𝑆 𝑒 , (90) driven wind of an NSM (Perego et al. 2014; Martin et al.
22

100
Final mass fractions MR-SNe NSM CC-SN
100

10 2 10 1
Mass fraction X

10 2

Mass fraction X
10 4
10 3

10 6 10 4

10 5

10 8
0 10 20 30 40 50 60 70 80 10 6
Mass number A 0 20 40 60 80 100 120 140
Mass number A

Figure 14. Final mass fractions of complete Si burning obtained

Figure 16. Final mass fractions after 1 Gyr for a trajectory of an
with a simple parametric model.
MR-SNe (Obergaulinger & Aloy 2017, Reichert et al. 2021), of the
neutrino-driven wind of an NSM (Perego et al. 2014, Martin et al.
2015), and of the neutrino-driven wind of a CC-SNe (Bliss et al.
Mass fraction
10 8 10 6 10 4 10 2 2018).

2015), and the neutrino-driven wind of a CC-SNe (Bliss et al.

Without neutrinos
45 2018). The final mass fractions are shown in Fig. 16.

40 5.1.6. Strong 𝑟-process

35 Calculating a full 𝑟-process is one of the most challeng-

ing nuclear reaction network calculations. Here we include
30 ∼ 6500 nuclei up to 337 Rg. The astrophysical host event of
the 𝑟-process is not fully understood yet. Very promising
25 candidates are NSMs, NSBH mergers, or MR-SNe. For these
Proton number

20 scenarios, we show the results of individual trajectories in

Fig. 17. These trajectories come from a variety of (M)HD
15 simulations and were presented in Winteler et al. (2012); Ko-
With neutrinos
45 robkin et al. (2012); Rosswog (2013); Piran et al. (2013); Wu
et al. (2016); Bovard et al. (2017); Obergaulinger & Aloy
40 (2017); Reichert et al. (2021); Obergaulinger & Aloy (2021);
35 Reichert et al. (2023).

30 5.2. Test scenarios

25 We have implemented a series of tests in order to monitor

the performance and consistency of WinNet. The tests cover
20 a range of numerical and physical scenarios, which we will
present in this section. Many of the tests are designed in a
15
10 20 30 40 50 way that an analytic calculation of the result is also possible.
Neutron number Furthermore, we implemented technical test cases such as
reading the initial composition, the correct reproduction of the
input thermodynamic conditions, and correct implementation
Figure 15. Mass fractions at 𝑡 = 1.8 × 103 s for one trajectory within of the different reaction rate formats.
the MR-SNe model 35OC-RO of Obergaulinger & Aloy (2017) and
Reichert et al. (2021). Upper panel: calculation without involving 5.2.1. 𝛽-decays
neutrino reactions. Lower panel: calculation using neutrino reac-
tions on nucleons as well as on heavier nuclei (Sieverding et al. A simple nucleosynthesis calculation is given by a 𝛽-decay.
2018). We tested the decay of neutrons to protons, as well as the decay
chain of 56 Ni. The results give an interesting insight into the
 23

100 100 Gear

10 1 Impl. Euler
Analytic
10 2
10 1

Mass fraction
Mass fraction X

10 3

10 4
10 2
10 5 MR-SNe 56Ni
NSM dyn. ejecta 56Co
10 6 NSM visc. ejecta 56Fe
NSBH dyn. ejecta 10 3
10 7 104 105 106 107 108
100 120 140 160 180 200 220 240 Time [s]
Mass number A

Figure 18. Decay of 56 Ni calculated with the Gear (solid line) and
Figure 17. Final mass fractions of various example trajectories. implicit Euler solver (dashed line). The analytic solution is shown
Within MR-SNe, the models used in Winteler et al. (2012), Ober- with the dotted lines.
gaulinger & Aloy (2017), Reichert et al. (2021), and Obergaulinger
& Aloy (2021), Reichert et al. (2023) are shown (red lines). For the
dynamic ejecta of an NSM (orange lines) we show the simulations
of Korobkin et al. (2012), Rosswog (2013), Piran et al. (2013, model
Kodama & Takahashi (1976)
ns10ns10), and Bovard et al. (2017). Furthermore, we illustrate the 10 1
Panov et al. (2001)
viscous ejecta of an NSM from the calculation of Wu et al. (2016).
Mumpower et al. (2020)
The dynamic ejecta of a NSBH merger from Korobkin et al. (2012),
10 2
Abundance

Rosswog (2013), Piran et al. (2013, model BH10) is shown as the

cyan line.
10 3

accuracy of the integration using an implicit Euler integration 10 4

scheme. We recall that this scheme does not have any error
estimate for the time step and the convergence criterion of 10 5

100 125 150 175 200 225 250 275 300

Í
the mass conservation (i.e., 𝑖 𝑋𝑖 = 1; Eq. (54)). While this
is common practice for calculations involving a large amount Mass number
of nuclei and reactions (e.g., Lippuner & Roberts 2017; Hix
& Thielemann 1999), within the tested decays it leads to
Figure 19. The 𝛽-delayed fission of 295 Am. Shown are the abun-
uncertainties. As an example, we show the time evolution of dances after 10 −2 s for three different fission fragment distributions.
56 Ni, 56 Co, and 56 Fe in Fig. 18. The discrepancies between

the implicit Euler solution and the analytic solution can be

lation time of 𝑡 = 10−2 s are shown in Fig. 19. The calculated
reduced by choosing adapted smaller time steps resulting from
abundance pattern deviates less than 1 % from the input fis-
smaller 𝜖Euler values (in the example, 𝜖 Euler = 10−1 was used;
sion fragment distributions.
see Eq. (57)). The example also shows the strength of the
adaptive time-step control within the Gear solver, which is
able to stay close to the analytic solution.
Another test is based on the 𝛽-delayed fission of 295 Am. 5.2.2. Equilibrium cases
Identical to a normal 𝛽-decay, we can calculate the decay of
Useful scenarios are cases were an equilibrium value is
this nucleus via
obtained. An equilibrium situation can be challenging for nu-
𝑌 (𝑡) = 𝑌0 𝑒 − 𝛼𝑡 , (94)
merical solvers, as constant abundances appear like a reaction
with the decay constant 𝛼. The products of this decay are time scale that is approaching infinity (e.g., Hix & Thielemann
determined by the fission fragment distribution that can be 1999; Feger 2011; Lippuner & Roberts 2017). In the follow-
calculated analytically, as in Kodama & Takahashi (1975) or ing, we present the case of an (n,𝛾)-(𝛾,n) equilibrium as well
Panov et al. (2001). Additionally, we include the fission frag- as equilibria obtained by electron- and positron-captures and
ment distribution of Mumpower et al. (2020) for 𝛽-delayed neutrino absorption.
and neutron-induced fission. This distribution spans a wide In the case of an (n,𝛾)-(𝛾,n) equilibrium between 64 Ni and
range of mass numbers. The different fragments for a simu- 65 Ni, the analytic solution of the equilibrium composition can
24

n 64Ni 65Ni Impl. Euler Gear Analytic

100 0.156 Electron and positron captures
0.155
Mass fraction

10 1 Gear 0.154
Impl. Euler 0.215 Neutrinos

Electron fraction
Time [s]
Analytic
10 2 0.214
0.213
106 Neutrinos + electron and positron captures
Timestep [s]

0.184
10 1
10 10 0.183
10 15 10 12 10 9 10 6 10 3 100 103
Time [s] 10 2 10 1 100 101 102 103

Figure 20. Upper panel: mass fractions of neutrons (blue), 64 Ni Figure 21. Electron fraction under hydrostatic conditions with
(orange), and 65 Ni (green) for the implicit Euler (dashed line) and 𝑇 = 30 GK and 𝜌 = 1010 g cm −3 . Upper panel: equilibrium case
Gear integration scheme (solid line). The analytic equilibrium so- when involving only electron- and positron-captures. Middle panel:
lution is shown with the black dotted lines. Lower panel: the time equilibrium case when involving only neutrinos with luminosities of
step of the implicit Euler and Gear integration schemes. 𝐿 𝜈 = 1052 erg s −1 , 𝐿 𝜈¯ = 5 × 1052 erg s −1 , and neutrino energies of
𝐸 𝜈 = 25.2 MeV as well as 𝐸 𝜈¯ = 31.5 MeV.
be derived as:
√︃ ons, neutrino absorption on nucleons, and a combination of
𝜆 𝛾,𝑛 − 𝜆2𝛾,𝑛 + 4𝜌 𝑁A ⟨𝜎𝜈⟩𝑛,𝛾 𝜆 𝛾,𝑛 /65 both. Similar to Just et al. (2022b), we can calculate the equi-
𝑌 (n) = (95) librium electron fractions for all three scenarios. Assuming
−2𝜌 𝑁A ⟨𝜎𝜈⟩𝑛,𝛾
only electron- and positron-captures, the equilibrium electron
𝑌 ( 64 Ni) = 𝑌 (n) (96) fraction for hydrostatic conditions can be obtained by solving:
𝑌 ( 65 Ni) = 1/65 − 𝑌 (n), (97)
𝜆 𝑒+ 𝑌 (n) − 𝜆 𝑒 − 𝑌 (p) = 0, (98)
For 𝑇 = 8 GK and 𝜌 = 109 g cm−3
and matter initially with the positron and electron-capture rate 𝜆 𝑒+ and 𝜆 𝑒 − , re-
consisting out of pure 𝑌 ( 65 Ni) (which introduced the fac- spectively. For a constant temperature of 30 GK and density
tor 1/65), we obtain 𝑌 (n) = 𝑌 ( 64 Ni) = 7.35175 × 10−3 of 1010 g cm−3 we obtain 𝑌e,em = 0.155. Both integration
and 𝑌 ( 65 Ni) = 8.03286 × 10−3 . While the integration with schemes obtain a great precision, with the Gear solver agree-
the Gear scheme results in an excellent agreement within ing within 0.004%, and the implicit Euler agreeing within
0.0015 %, the time step within the implicit Euler becomes 0.076%. The implicit Euler integration scheme shows again
very small. This leads to numerical instabilities and a large some numerical noise (upper panel in Fig. 21) and the result
deviation from the analytic solution after 103 s (see Fig. 20). is therefore slightly worse compared to the Gear integration
This instability is unlikely to be resolved by more restrictive scheme. For the scenario with only neutrinos irradiating the
time steps in the implicit Euler scheme as the time step is matter, we can similarly calculate the equilibrium electron
based of changes in the thermodynamic conditions or abun- fraction:
dances. Since both are static, the scheme will always attempt 𝜆 𝜈𝑒 𝑌 (n) − 𝜆 𝜈¯𝑒 𝑌 (p) = 0, (99)
very large (possibly too large) time steps. This continues until
with the neutrino and antineutrino cross sections 𝜆 𝜈𝑒 and 𝜆 𝜈¯𝑒 ,
large errors have been accumulated and the solution diverges.
respectively. Assuming matter located at a radius of 50 km,
On the other hand, the Gear solver estimates an integration
irradiated by neutrino luminosities of 𝐿 𝜈 = 1052 erg s−1 , 𝐿 𝜈¯ =
error, independent on changes in conditions or abundances.
5 × 1052 erg s−1 , and neutrino energies of 𝐸 𝜈 = 25.2 MeV as
As a consequence, the solution is more stable.
well as 𝐸 𝜈¯ = 31.5 MeV we obtain 𝑌e,abs = 0.214. The final
Another equilibrium test scenario is given by the equilib-
values of both integration schemes agree within 5 × 10−8 %
rium of electron-, positron-, and neutrino-captures. These
(middle panel of Fig. 21). Combining electron-, positron-,
equilibria are important to understand the initial electron
and neutrino-captures, the equilibrium electron fraction can
fraction in 𝑟-process calculations. In the following, we in-
be obtained by solving
vestigate the situation of the equilibrium electron fraction
when only considering electron-/positron-captures on nucle- (𝜆 𝜈𝑒 + 𝜆 𝑒+ )𝑌 (n) − (𝜆 𝜈¯𝑒 + 𝜆 𝑒 − )𝑌 (p) = 0. (100)
 25

A Z YWinNet ΔSkyNet ΔReNet ΔXNet SkyNet WinNet

[10 −2 %] [10 −3 %] [10 −1 %] SkyNet (db) WinNet (db)
100
4 2 4.01 × 10 −09 0.82 0.13 0.25
12 6 4.04 × 10 −18 0.53 0.08 0.16
16 8 1.55 × 10 −16 8.06 1.27 2.44 10 2
20 10 1.89 × 10 −19 7.23 1.14 2.19

Mass fraction
24 12 1.14 × 10 −14 6.61 1.04 2.00
28 14 8.14 × 10 −09 5.79 0.91 1.75 10 4

32 16 4.52 × 10 −08 4.96 0.78 1.50

36 18 7.54 × 10 −09 4.13 0.65 1.25 10 6
40 20 5.78 × 10 −07 3.31 0.52 1.00
44 22 2.89 × 10 −09 2.48 0.39 0.75
48 24 3.23 × 10 −07 1.65 0.26 0.50 10 8
0 20 40 60 80 100 120
52 26 7.13 × 10 −05 0.82 0.13 0.25 Mass number
56 28 1.78 × 10 −02 0.003 0.001 0.001
Table 3. Final abundances for hydrostatic Carbon-Oxgen test case.
Figure 22. Composition of an X-ray burst (Schatz et al. 2001) after
Columns 4-6 show the deviation compared to the results of SkyNet,
103 s. The result is shown for WinNet (orange lines) and SkyNet
ReNet, and XNet, respectively.
(blue lines) with (dashed lines) and without (solid lines) calculating
reverse reactions via detailed balance.
For the conditions assumed here, we obtain 𝑌e,𝛽 = 0.1835,
which only deviates by 0.014 % from the equilibrium value did not tune the specific numerical parameters used in the
(lower panel of Fig. 21). different codes. More restrictive time steps can therefore lead
Another equilibrium case is given by NSE (Section 2.3). to an even better agreement.
We tested that the transition from the NSE region to the To test the implementation of detailed balance (Section 2.2),
network region is consistent. Therefore, we calculated the we repeated the calculation performed in Lippuner & Roberts
NSE composition for 𝑇 = 7 GK, 𝜌 = 107 g cm−3 , and 𝑌𝑒 = (2017) with SkyNet. We calculate the nucleosynthesis of a
0.5 with and without screening. In addition, we calculated trajectory of an X-ray burst from Schatz et al. (2001). The
the mass fractions after 102 s when starting with neutrons result of four calculations is shown in Fig. 22. There, we
and protons only, using the same hydrostatic conditions and use the Reaclib v2.2 and calculate the nucleosynthesis with
strong reactions only. This system should also approach NSE. WinNet using the reverse rates as given by Reaclib (solid or-
Again, we calculate the abundances with and without electron ange line). Moreover, we use SkyNet with reverse rates from
screening corrections (see Fig. 2). As a consequence of the Reaclib (solid blue line). Additionally, we calculate the same
previous outlined tests, we only calculated the test with the trajectory, but using reverse rates calculated via detailed bal-
Gear integration method. ance using the Q-value from the mass excess provided within
Reaclib (within the winvn file, dashed lines). Both networks
5.2.3. Other tests agree very well for both cases. The impact of using detailed
balance rates with masses from the winvn in contrast to Rea-
If the nuclear reaction network is sufficiently large, deriving
clib reverse rates seems to be larger in SkyNet especially for
an analytic expression for the solution is often not possible
the smaller mass numbers ∼ 50. However, for both networks,
anymore. In these cases, it is beneficial to compare the result
there is also a distinct feature visible at 𝐴 ∼ 85.
with other nuclear reaction networks.
We calculate a case of hydrostatic carbon-Oxygen burning
with 𝜌 = 109 g cm−3 and a temperature of 3 GK for 1012 s. 6. SUMMARY AND CONCLUSION
The initial composition consisted of 𝑋 ( 12 C) = 𝑋 ( 16 O) = 0.5. We have summarized the fundamentals of nuclear reaction
In total we involve 13 nuclei in the calculation. We compare networks. The implementation was demonstrated with the
the final abundances of WinNet (using Gears integration single-zone nuclear reaction network code WinNet.
method) with the results of the nuclear reaction networks We outlined the differential equations that underlie every
SkyNet (Lippuner & Roberts 2017), ReNet (Navó et al. nuclear reaction network code. Additionally, we presented
2023), and XNet (Hix & Thielemann 1999). The final abun- two implicit numerical techniques to solve these equations,
dances of WinNet deviate by less than 1% to all other reaction the implicit Euler and Gear’s integration scheme.
networks (Tab. 3). The abundant nucleus 56 Ni even agrees A mandatory ingredient is also the set of reaction rates.
with a maximum deviation of 10−4 % only. We note that we The reaction rates can originate from different databases with
26

varying parameterizations. Hereby, one should ensure that the the code. Due to the modular structure, it is also easy to
same underlying nuclear physic inputs such as mass models change the included reactions. Additionally, large effort has
are used. We described the reaction rate formats that are sup- been undertaken to supply understandable error messages. To
ported by WinNet, namely the Reaclib reaction rate database, give an example, if an input parameter is misspelled, the error
a format for 𝛽-delayed neutron emission, tabulated rates, the- message contains not only that this parameter does not exist,
oretical 𝛽+ , 𝛽 − , electron-capture and positron-capture rates, but also points to the most similar existing parameter.
neutrino reactions, and fission reactions. When deciding for or against favoring the usage of WinNet
All these different reaction sources get a different priority over other publicly available reaction networks, one should
assigned, and rates that appear in more than one source are keep in mind the advantages and disadvantages for carrying
replaced by the rate with the highest priority. This priority is out the desired task. An obvious point to make here is the
chosen arbitrarily without any estimate of the quality of the code language. Users that want to make changes in the code
rate, and there could be still some action required if a user and that are more familiar with C or C++ may feel more
wants to use specific reaction rates. comfortable with using NucNet (Meyer & Adams 2007) or
WinNet is further able to calculate detailed balance reac- SkyNet (Lippuner & Roberts 2017) rather than the Fortran
tions on-the-fly, which can be useful especially for tabulated 90 written codes of Torch (Paxton et al. 2015), XNet (Hix
rates, where a tabulation of the reverse reactions could break & Thielemann 1999), or WinNet. For applications that re-
the detailed balance principle. If included, the detailed bal- quire including the reaction network into a hydrodynamical
ance reactions will replace all reverse reactions in the other code, the usage of Torch or XNet may be favored over Win-
reaction rate sources. Net. While it is not impossible to include WinNet into a
This is also especially useful when implementing new re- hydrodynamical code, there is more experience with Torch
action rates for which the reverse rates may not always have or XNet as both have already been used in hydrodynamical
been published. As an example, deBoer et al. (2017) pub- simulations. Furthermore, there has been more effort in par-
lished the reaction rate 12 C(𝛼, 𝛾) 16 O. While the forward rate allelizing and optimizing XNet especially when calculating
can easily be changed as they give Reaclib parameters (Sec- abundances for more than one zone. Since the calculation
tion 4.2.1) and tabulated values (Section 4.2.3), the reverse of the ejecta of astrophysical events often relies on many
reaction should also be consistently changed (Section 2.2). independent tracer particles, the parallelization of WinNet
Instead of calculating this rate by hand, within WinNet one relies on executing many instances of the reaction network
can enable a parameter to calculate the reverse reaction inter- for different tracer particles. No effort has been made in
nally. parallelizing the calculation of a single tracer particle, and
All charged particle reaction rates can be further altered by a single instance of WinNet should always be executed on
electron screening. This correction is implemented with a one CPU only. Similar to XNet and in contrast to SkyNet,
multiplicative factor to the reaction rates. WinNet optionally performs an initialization step to bring the
We presented the energy feedback from nuclear reactions included reaction rates into an advantageous shape to mini-
onto the temperature, which is implemented in the form of an mize the cost of reading them in. Especially when running
operator splitting method. many tracers, this can reduce the computational cost of the
Finally, we introduced simple examples and test cases to initialization step. In post-processed applications where neu-
demonstrate the reliability of the reaction network code Win- trinos play a major role, WinNet definitely has advantages
Net. Using these test cases, we analyzed the advantages as it is, to our knowledge, the only public code that is able
and disadvantages of the different implemented numerical to use publicly available neutrino reactions on heavier nuclei
integration methods. We conclude that hydrostatic and equi- (Section 4.2.4)19. Furthermore, for conditions that span a
librium conditions are often more efficient and precise with large range of temperatures and densities, the scheme to use
the Gear integration method. More complex rapidly varying and replace theoretical ec-/pc-/𝛽+ -/𝛽 − -reactions with exper-
thermodynamic conditions are more efficient with the implicit imental half-lives contained in the Jina REACLIB database
Euler integration method. (Section 4.2.5) is an advantage of WinNet. Regarding cal-
In addition to the reliability, a large focus during the devel- culations of very neutron-rich environments, to our knowl-
opment of WinNet was the usability. The code provides an edge, Torch was never run in the context of the 𝑟-process.
easy interface to the user by a simple parameter file. Addition- It is possible to include fission reactions and fragment dis-
ally, comments are written entirely in a doxygen18 conform tributions into XNet (as done in, e.g., Lippuner & Roberts
format, and the documentation can be accessed along with
19 We note that XNet can also contain neutrino reactions on heavier nuclei;
18 https://www.doxygen.nl/index.html the default rate tables are however not publicly available, and one would
have to convert the format of the public rate tables.
 27

2017). However, including more complex fragment distribu- parts credit, the corresponding work should be cited when
tions with hundreds of fission fragments might be challeng- someone makes use of a later added feature.
ing and not possible without code changes. To date, also
SkyNet includes only a relatively simple fragment distribu- We want to thank M. A. Aloy, F. Montes, M. Obergaulinger,
tion with mostly only two fragments plus neutron emission. T. Psaltis, H. Schatz, A. Sieverding, and M. Ugliano for many
If one wants to carry out a study with more complex fission beneficial discussions. We further thank all of the people that
fragments, WinNet could be a better choice, as an arbitrary made example trajectories publicly available and L. Bovard,
amount of fragments can be included (see Section 4.2.7). Ad- R. Fernández, M. Obergaulinger, and H. Schatz for giving
ditionally, we included parameterized 𝛼-decays into WinNet their consent to include their trajectories along with WinNet.
which is an additional, even though possibly small, effort to Furthermore, we thank F. Timmes for making many useful
include into SkyNet or XNet as well. For testing the impact tools publicly available and for allowing us to use them. We
of newly measured reaction rates in different environments, additionally want to thank the referee for providing useful
WinNet is also a good choice, since we consider it relatively suggestions that helped to improve the manuscript and cer-
easy to exchange reaction rates, but also because WinNet tain aspects of WinNet itself. M.R. acknowledges support
already comes with an extensive set of example cases where from the grants FJC2021-046688-I and PID2021-127495NB-
the impact of certain rates can be directly tested. Also, for I00, funded by MCIN/AEI/10.13039/501100011033 and by
storage critical applications, WinNet includes a very flexible the European Union ”NextGenerationEU” as well as ”ESF
way of turning output on and off. This is not easily possible Investing in your future”. Additionally, he acknowledges sup-
within SkyNet or Torch without touching the code itself. port from the Astrophysics and High Energy Physics program
For example, if one wants to know only the abundances af- of the Generalitat Valenciana ASFAE/2022/026 funded by
ter 1 day, it is possible within WinNet to exclusively output MCIN and the European Union NextGenerationEU (PRTR-
the abundances at this time. Furthermore, WinNet is able C17.I1). A.A., G.M.P., J.K., and M.J. acknowledge sup-
to output either ascii, hdf5, or both files, and we therefore port by the Deutsche Forschungsgemeinschaft (DFG, German
consider the code specifically user and beginner friendly. In Research Foundation) – Project-ID 279384907 - SFB 1245
contrast, SkyNet necessarily needs to have hdf5 packages and the State of Hessen within the Research Cluster ELE-
installed. Ultimately, for numerical studies where the avail- MENTS (Project ID 500/10.006). A.A., J.K., and M.J. ad-
ability of different numerical solvers is desirable, WinNet ditionally acknowledge support from the European Research
or XNet are favored over SkyNet, which only includes a Council under grant EUROPIUM-667912. G.M.P. acknowl-
backward Euler integration. On the other hand, XNet and edges support by the ERC under the European Union’s Hori-
SkyNet include more possibilities for the exploration of ma- zon 2020 research and innovation program (ERC Advanced
trix inversion packages, as WinNet includes only the sparse grant KILONOVA No. 885281). O.K. was supported by the
PARDISO solver. We also briefly tested the performance of US Department of Energy through the Los Alamos National
WinNet in comparison to SkyNet for an 𝑟-process example. Laboratory (LANL). LANL is operated by Triad National
In this test, both codes were similarly fast with WinNet being Security, LLC, for the National Nuclear Security Administra-
slightly faster. However, to get a firmer and more quantitative tion of the U.S.DOE (contract No. 89233218CNA000001).
comparison of the performance, a more detailed investigation This work is authorized for unlimited release under LA-UR-
that covers additional numerical parameters and astrophysical 23-25461. C.F. acknowledges support from the by United
conditions would be necessary. States Department of Energy, Office of Science, Office of
With this work, WinNet will be fully public and available Nuclear Physics (award No. DE-FG02-02ER41216). R.H.
for download at GitHub20 and Zenodo21 (Reichert et al. acknowledges support from the World Premier International
2023). This includes not only the code, but also all example Research Centre Initiative (WPI Initiative), MEXT, Japan;
and test cases. If you use them, please cite the according the IReNA AccelNet Network of Networks, supported by the
publications that can be found in the documentation. National Science Foundation under grant No. OISE-1927130
With future works, we plan to extend WinNet by adding and ChETEC-INFRA (grant No. 101008324) supported by
more features. As an example, a sensitivity study that uses the European Union’s Horizon 2020 research and innova-
WinNet may extend it by the scripts and code extensions to tion program. This publication benefited highly from col-
perform this task. To give the individual authors of these laborations and exchange within the European Cost Action
CA16117 ”Chemical Evolution as Tracers of the Evolution
of the Cosmos” (ChETEC) and the ”International Research
20 https://github.com/nuc-astro/WinNet
Network for Nuclear Astrophysics” (IReNA).
21 https://zenodo.org/record/8220549
28

Software: Matplotlib (Hunter 2007), Numpy (Harris et al. et al. 2023), XNet (Hix & Thielemann 1999), SkyNet (Lip-
2020), Scipy (Virtanen et al. 2020), Quadpack (Piessens et al. puner & Roberts 2017), PARDISO (Schenk & Gärtner 2004).
1983), Timmes EOS (Timmes & Arnett 1999), ReNet (Navó

REFERENCES
Aikawa, M., Arnould, M., Goriely, S., Jorissen, A., & Takahashi, Bravo, E. 2020, MNRAS, 494, 3037, doi: 10.1093/mnras/staa910
K. 2005, A&A, 441, 1195, doi: 10.1051/0004-6361:20052944 Bravo, E., & Garcı́a-Senz, D. 1999, MNRAS, 307, 984,
Aloy, M. Á., & Obergaulinger, M. 2021, MNRAS, 500, 4365, doi: 10.1046/j.1365-8711.1999.02694.x
doi: 10.1093/mnras/staa3273 Brown, B. A. 1992, PhRvC, 46, 811,
Alpher, R. A., Bethe, H., & Gamow, G. 1948, Physical Review, 73, doi: 10.1103/PhysRevC.46.811
803, doi: 10.1103/PhysRev.73.803 Brown, D. A., Chadwick, M. B., Capote, R., et al. 2018, Nuclear
Arcones, A., Martı́nez-Pinedo, G., Roberts, L. F., & Woosley, S. E. Data Sheets, 148, 1, doi: 10.1016/j.nds.2018.02.001
2010, A&A, 522, A25, doi: 10.1051/0004-6361/201014276
Brown, P. N., Byrne, G. D., & Hindmarsh, A. C. 1989, SIAM
Arnett, D. 1996, Supernovae and Nucleosynthesis: An Investigation
Journal on Scientific and Statistical Computing, 10, 1038,
of the History of Matter from the Big Bang to the Present
doi: 10.1137/0910062
Arnett, W. D. 1969, Astrophys. Space. Sci., 5, 180,
Broyden, C. G. 1965, Mathematics of Computation, 19, 577
doi: 10.1007/BF00650291
Bruenn, S. W. 1986, ApJS, 62, 331, doi: 10.1086/191143
—. 1977, ApJS, 35, 145, doi: 10.1086/190472
Burrows, A. 2013, Rev. Mod. Phys., 85, 245,
Arnett, W. D., & Truran, J. W. 1969, ApJ, 157, 339,
doi: 10.1086/150072 doi: 10.1103/RevModPhys.85.245
Arnett, W. D., Truran, J. W., & Woosley, S. E. 1971, ApJ, 165, 87, Burrows, A., Reddy, S., & Thompson, T. A. 2006, NuPhA, 777,
doi: 10.1086/150878 356, doi: 10.1016/j.nuclphysa.2004.06.012
Arnett, W. D., Meakin, C., Hirschi, R., et al. 2019, ApJ, 882, 18, Busso, M., Vescovi, D., Palmerini, S., Cristallo, S., &
doi: 10.3847/1538-4357/ab21d9 Antonuccio-Delogu, V. 2021, ApJ, 908, 55,
Arnould, M. 1976, A&A, 46, 117 doi: 10.3847/1538-4357/abca8e
Arnould, M., Norgaard, H., Thielemann, F. K., & Hillebrandt, W. Byrne, G., & Hindmarsh, A. 1975, ACM Transactions on
1980, ApJ, 237, 931, doi: 10.1086/157940 Mathematical Software (TOMS), 1, 71,
Aver, E., Olive, K. A., & Skillman, E. D. 2010, JCAP, 2010, 003, doi: 10.1145/355626.355636
doi: 10.1088/1475-7516/2010/05/003 Cardall, C. Y., & Fuller, G. M. 1997, ApJL, 486, L111,
Bania, T. M., Rood, R. T., & Balser, D. S. 2002, Nature, 415, 54, doi: 10.1086/310838
doi: 10.1038/415054a Cescutti, G. 2022, Main s-process, 1.2.1, Zenodo,
Barnes, J., & Metzger, B. D. 2022, ApJL, 939, L29, doi: 10.5281/zenodo.6474686
doi: 10.3847/2041-8213/ac9b41 Cescutti, G., Hirschi, R., Nishimura, N., et al. 2018, MNRAS, 478,
Benz, W., Hills, J. G., & Thielemann, F. K. 1989, ApJ, 342, 986, 4101, doi: 10.1093/mnras/sty1185
doi: 10.1086/167656
Clayton, D. 1968, Principles of stellar evolution and
Bisterzo, S., Travaglio, C., Wiescher, M., Käppeler, F., & Gallino,
nucleosynthesis: with a new preface (University of Chicago
R. 2017, ApJ, 835, 97, doi: 10.3847/1538-4357/835/1/97
Press)
Bliss, J., Arcones, A., Montes, F., & Pereira, J. 2020, PhRvC, 101,
Coc, A., & Vangioni, E. 2017, International Journal of Modern
055807, doi: 10.1103/PhysRevC.101.055807
Physics E, 26, 1741002, doi: 10.1142/S0218301317410026
Bliss, J., Arcones, A., & Qian, Y.-Z. 2018, ApJ, 866, 105,
Cooke, R. J., Pettini, M., & Steidel, C. C. 2018, ApJ, 855, 102,
doi: 10.3847/1538-4357/aade8d
doi: 10.3847/1538-4357/aaab53
Boesgaard, A. M., & Steigman, G. 1985, ARA&A, 23, 319,
doi: 10.1146/annurev.aa.23.090185.001535 Curtis, S., Ebinger, K., Fröhlich, C., et al. 2019, ApJ, 870, 2,
Bollig, R., Janka, H. T., Lohs, A., et al. 2017, PhRvL, 119, 242702, doi: 10.3847/1538-4357/aae7d2
doi: 10.1103/PhysRevLett.119.242702 Cyburt, R. H., Fields, B. D., Olive, K. A., & Yeh, T.-H. 2016, Rev.
Bovard, L., Martin, D., Guercilena, F., et al. 2017, PhRvD, 96, Mod. Phys., 88, 015004, doi: 10.1103/RevModPhys.88.015004
124005, doi: 10.1103/PhysRevD.96.124005 Cyburt, R. H., Amthor, A. M., Ferguson, R., et al. 2010, The
Bovard, L., & Rezzolla, L. 2017, Classical and Quantum Gravity, Astrophysical Journal Supplement Series, 189, 240,
34, 215005, doi: 10.1088/1361-6382/aa8d98 doi: 10.1088/0067-0049/189/1/240
 29

Dan, M., Guillochon, J., Brüggen, M., Ramirez-Ruiz, E., & Fujimoto, S.-i., Nishimura, N., & Hashimoto, M.-a. 2008, ApJ,
Rosswog, S. 2015, MNRAS, 454, 4411, 680, 1350, doi: 10.1086/529416
doi: 10.1093/mnras/stv2289 Fuller, G. M., Fowler, W. A., & Newman, M. J. 1982, ApJS, 48,
Dardelet, L., Ritter, C., Prado, P., et al. 2015, arXiv e-prints, 279, doi: 10.1086/190779
arXiv:1505.05500. https://arxiv.org/abs/1505.05500 —. 1985, ApJ, 293, 1, doi: 10.1086/163208
deBoer, R. J., Görres, J., Wiescher, M., et al. 2017, Rev. Mod. Garcia-Senz, D., Cabezon, R. M., Arcones, A., Relano, A., &
Phys., 89, 035007, doi: 10.1103/RevModPhys.89.035007 Thielemann, F. K. 2013, MNRAS, 436, 3413,
Denissenkov, P. A., Herwig, F., Woodward, P., et al. 2019, doi: 10.1093/mnras/stt1821
MNRAS, 488, 4258, doi: 10.1093/mnras/stz1921 Garcı́a-Senz, D., Cabezón, R. M., Domı́nguez, I., & Thielemann,
Descouvemont, P., Adahchour, A., Angulo, C., Coc, A., & F. K. 2016, ApJ, 819, 132, doi: 10.3847/0004-637X/819/2/132
Vangioni-Flam, E. 2004, Atomic Data and Nuclear Data Tables, Gear, C. W. 1971, Commun. ACM, 14, 176,
88, 203, doi: 10.1016/j.adt.2004.08.001 doi: 10.1145/362566.362571
Dewitt, H. E., Graboske, H. C., & Cooper, M. S. 1973, ApJ, 181, Ghosh, S., Wolfe, N., & Fröhlich, C. 2022, ApJ, 929, 43,
439, doi: 10.1086/152061 doi: 10.3847/1538-4357/ac4d20
Dillmann, I., Heil, M., Käppeler, F., et al. 2006, in American Gil-Pons, P., Doherty, C. L., Gutiérrez, J. L., et al. 2018, PASA, 35,
Institute of Physics Conference Series, Vol. 819, Capture e038, doi: 10.1017/pasa.2018.42
Giuliani, S. A., Martı́nez-Pinedo, G., & Robledo, L. M. 2018a,
Gamma-Ray Spectroscopy and Related Topics, ed. A. Woehr &
PhRvC, 97, 034323, doi: 10.1103/PhysRevC.97.034323
A. Aprahamian, 123–127, doi: 10.1063/1.2187846
Giuliani, S. A., Martı́nez-Pinedo, G., & Robledo, L. M. 2018b, in
Doherty, C. L., Gil-Pons, P., Siess, L., & Lattanzio, J. C. 2017,
Journal of Physics Conference Series, Vol. 940, Journal of
PASA, 34, e056, doi: 10.1017/pasa.2017.52
Physics Conference Series, 012013,
Dong, T., & Ren, Z. 2005, European Physical Journal A, 26, 69,
doi: 10.1088/1742-6596/940/1/012013
doi: 10.1140/epja/i2005-10142-y
Giuliani, S. A., Martı́nez-Pinedo, G., Wu, M.-R., & Robledo, L. M.
Eggenberger, P., Ekström, S., Georgy, C., et al. 2021, A&A, 652,
2020, PhRvC, 102, 045804, doi: 10.1103/PhysRevC.102.045804
A137, doi: 10.1051/0004-6361/202141222
Goriely, S., Bauswein, A., & Janka, H.-T. 2011, ApJL, 738, L32,
Eichler, M., Arcones, A., Kelic, A., et al. 2015, ApJ, 808, 30,
doi: 10.1088/2041-8205/738/2/L32
doi: 10.1088/0004-637X/808/1/30
Goriely, S., Hilaire, S., Koning, A. J., Sin, M., & Capote, R. 2009,
Erler, J., Langanke, K., Loens, H., Martinez-Pinedo, G., &
PhRvC, 79, 024612, doi: 10.1103/PhysRevC.79.024612
Reinhard, P.-G. 2012, 85, 025802,
Görres, J., Wiescher, M., & Thielemann, F.-K. 1995, PhRvC, 51,
doi: 10.1103/PhysRevC.85.025802
392, doi: 10.1103/PhysRevC.51.392
Feger, E. D. 2011, PhD thesis, University of Tennessee, Tennessee. Gronow, S., Côté, B., Lach, F., et al. 2021, A&A, 656, A94,
https://trace.tennessee.edu/utk graddiss/1048 doi: 10.1051/0004-6361/202140881
Fields, B. D. 2011, Annual Review of Nuclear and Particle Science, Halevi, G., & Mösta, P. 2018, MNRAS, 477, 2366,
61, 47, doi: 10.1146/annurev-nucl-102010-130445 doi: 10.1093/mnras/sty797
Fields, B. D., & Olive, K. A. 2022, JCAP, 2022, 078, Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020,
doi: 10.1088/1475-7516/2022/10/078 Nature, 585, 357, doi: 10.1038/s41586-020-2649-2
Fowler, W. A. 1974, QJRAS, 15, 82 Harris, J. A., Hix, W. R., Chertkow, M. A., et al. 2017, ApJ, 843, 2,
Fowler, W. A., Caughlan, G. R., & Zimmerman, B. A. 1967, doi: 10.3847/1538-4357/aa76de
ARA&A, 5, 525, doi: 10.1146/annurev.aa.05.090167.002521 Heger, A., Fryer, C. L., Woosley, S. E., Langer, N., & Hartmann,
Freiburghaus, C., Rosswog, S., & Thielemann, F.-K. 1999, ApJL, D. H. 2003, ApJ, 591, 288, doi: 10.1086/375341
525, L121, doi: 10.1086/312343 Heger, A., & Woosley, S. E. 2010, ApJ, 724, 341,
Frischknecht, U., Hirschi, R., Pignatari, M., et al. 2016, MNRAS, doi: 10.1088/0004-637X/724/1/341
456, 1803, doi: 10.1093/mnras/stv2723 Hillebrandt, W., Kromer, M., Röpke, F. K., & Ruiter, A. J. 2013,
Fröhlich, C., Martı́nez-Pinedo, G., Liebendörfer, M., et al. 2006a, Frontiers of Physics, 8, 116, doi: 10.1007/s11467-013-0303-2
Physical Review Letters, 96, 142502, Hirschi, R., Meynet, G., & Maeder, A. 2004, A&A, 425, 649,
doi: 10.1103/PhysRevLett.96.142502 doi: 10.1051/0004-6361:20041095
Fröhlich, C., Hauser, P., Liebendörfer, M., et al. 2006b, ApJ, 637, Hix, W. R., Khokhlov, A. M., Wheeler, J. C., & Thielemann, F.-K.
415, doi: 10.1086/498224 1998, ApJ, 503, 332, doi: 10.1086/305968
Fryxell, B., Mueller, E., & Arnett, D. 1991, ApJ, 367, 619, Hix, W. R., & Meyer, B. S. 2006, Nuclear Physics A, 777, 188,
doi: 10.1086/169657 doi: 10.1016/j.nuclphysa.2004.10.009
30

Hix, W. R., Parete-Koon, S. T., Freiburghaus, C., & Thielemann, Karakas, A. I., & Lattanzio, J. C. 2014, PASA, 31, e030,
F.-K. 2007, ApJ, 667, 476, doi: 10.1086/520672 doi: 10.1017/pasa.2014.21
Hix, W. R., & Thielemann, F.-K. 1999, Journal of Computational Karakas, A. I., & Lugaro, M. 2016, ApJ, 825, 26,
and Applied Mathematics, 109, 321 doi: 10.3847/0004-637X/825/1/26
Hoffman, R. D., Woosley, S. E., & Qian, Y.-Z. 1997, ApJ, 482, Kawano, L., Schramm, D., & Steigman, G. 1988, ApJ, 327, 750,
951, doi: 10.1086/304181 doi: 10.1086/166232
Hoffman, R. D., Woosley, S. E., Weaver, T. A., Rauscher, T., & Kelic, A., Valentina Ricciardi, M., & Schmidt, K.-H. 2009, arXiv
Thielemann, F. K. 1999, ApJ, 521, 735, doi: 10.1086/307568 e-prints, arXiv:0906.4193. https://arxiv.org/abs/0906.4193
Höflich, P., Wheeler, J. C., & Thielemann, F. K. 1998, ApJ, 495, Khokhlov, A., Mueller, E., & Hoeflich, P. 1993, A&A, 270, 223
617, doi: 10.1086/305327 Khokhlov, A. M. 1991, A&A, 245, 114
Holmbeck, E. M., Sprouse, T. M., Mumpower, M. R., et al. 2019, Khuyagbaatar, J. 2020, NuPhA, 1002, 121958,
ApJ, 870, 23, doi: 10.3847/1538-4357/aaefef doi: 10.1016/j.nuclphysa.2020.121958
Holmes, J. A., Woosley, S. E., Fowler, W. A., & Zimmerman, B. A. Kippenhahn, R., Weigert, A., & Weiss, A. 2013, Stellar Structure
1976, At. Data Nucl. Data Tables, 18, 305, and Evolution, 2nd edn. (Berin Heidelberg: Springer),
doi: 10.1016/0092-640X(76)90011-5 doi: 10.1007/978-3-642-30304-3
Horowitz, C. J. 2002, PhRvD, 65, 043001, Kobayashi, C., Karakas, A. I., & Lugaro, M. 2020, ApJ, 900, 179,
doi: 10.1103/PhysRevD.65.043001 doi: 10.3847/1538-4357/abae65
Howard, W. M., & Möller, P. 1980, 25, 219, Kodama, T., & Takahashi, K. 1975, NuPhA, 239, 489,
doi: 10.1016/0092-640X(80)90005-4 doi: 10.1016/0375-9474(75)90381-4
Hunter, J. D. 2007, Computing in Science & Engineering, 9, 90, Koike, O., Hashimoto, M.-a., Kuromizu, R., & Fujimoto, S.-i.
doi: 10.1109/MCSE.2007.55 2004, ApJ, 603, 242, doi: 10.1086/381354
Iben, I., J., & Tutukov, A. V. 1984, ApJS, 54, 335, Koning, A. J., Rochman, D., Sublet, J. C., et al. 2019, Nuclear Data
doi: 10.1086/190932 Sheets, 155, 1, doi: 10.1016/j.nds.2019.01.002
Ichimaru, S. 1993, Rev. Mod. Phys., 65, 255, Korobkin, O., Rosswog, S., Arcones, A., & Winteler, C. 2012,
doi: 10.1103/RevModPhys.65.255 MNRAS, 426, 1940, doi: 10.1111/j.1365-2966.2012.21859.x
Iliadis, C. 2015, Nuclear Physics of Stars (Wiley-VCH Verlag Kostka, M., Koning, N., Shand, Z., Ouyed, R., & Jaikumar, P.
GmbH & Co. KGaA), doi: 10.1002/9783527692668 2014, A&A, 568, A97, doi: 10.1051/0004-6361/201322887
Iliadis, C., Champagne, A., José, J., Starrfield, S., & Tupper, P. Kotake, K., Takiwaki, T., Suwa, Y., et al. 2012, Advances in
2002, ApJS, 142, 105, doi: 10.1086/341400 Astronomy, 2012, 428757, doi: 10.1155/2012/428757
Itoh, N., Hayashi, H., Nishikawa, A., & Kohyama, Y. 1996, ApJS, Kratz, K. L., Farouqi, K., Mashonkina, L. I., & Pfeiffer, B. 2008,
102, 411, doi: 10.1086/192264 NewAR, 52, 390, doi: 10.1016/j.newar.2008.06.015
Janka, H.-T., Melson, T., & Summa, A. 2016, Annual Rev. of Kravchuk, P. A., & Yakovlev, D. G. 2014, PhRvC, 89, 015802,
Nuclear and Particle Science, 66, 341, doi: 10.1103/PhysRevC.89.015802
doi: 10.1146/annurev-nucl-102115-044747 Kullmann, I., Goriely, S., Just, O., et al. 2022, MNRAS, 510, 2804,
Jiang, J.-A., Doi, M., Maeda, K., et al. 2017, Nature, 550, 80, doi: 10.1093/mnras/stab3393
doi: 10.1038/nature23908 Kullmann, I., Goriely, S., Just, O., Bauswein, A., & Janka, H. T.
Jose, J. 2016, Stellar Explosions: Hydrodynamics and 2023, MNRAS, 523, 2551, doi: 10.1093/mnras/stad1458
Nucleosynthesis (Boca Raton: CRC Press), doi: 10.1201/b19165 Kushnir, D., Waxman, E., & Chugunov, A. I. 2019, MNRAS, 486,
José, J. 2022, Nova outburst, 1.1.1, Zenodo, 449, doi: 10.1093/mnras/stz904
doi: 10.5281/zenodo.6474694 Lach, F., Callan, F. P., Bubeck, D., et al. 2022, A&A, 658, A179,
José, J., & Hernanz, M. 1998, ApJ, 494, 680, doi: 10.1086/305244 doi: 10.1051/0004-6361/202141453
José, J., Hernanz, M., Amari, S., Lodders, K., & Zinner, E. 2004, Langanke, K., & Kolbe, E. 2002, Atomic Data and Nuclear Data
ApJ, 612, 414, doi: 10.1086/422569 Tables, 82, 191, doi: 10.1006/adnd.2002.0883
Just, O., Aloy, M. A., Obergaulinger, M., & Nagataki, S. 2022a, Langanke, K., & Martı́nez-Pinedo, G. 2001, Atomic Data and
ApJL, 934, L30, doi: 10.3847/2041-8213/ac83a1 Nuclear Data Tables, 79, 1, doi: 10.1006/adnd.2001.0865
Just, O., Goriely, S., Janka, H. T., Nagataki, S., & Bauswein, A. Leung, S.-C., & Nomoto, K. 2018, ApJ, 861, 143,
2022b, MNRAS, 509, 1377, doi: 10.1093/mnras/stab2861 doi: 10.3847/1538-4357/aac2df
Kaiser, E. A., Hirschi, R., Arnett, W. D., et al. 2020, MNRAS, 496, Leung, S.-C., Nomoto, K., & Suzuki, T. 2020, ApJ, 889, 34,
1967, doi: 10.1093/mnras/staa1595 doi: 10.3847/1538-4357/ab5d2f
 31

Limongi, M., & Chieffi, A. 2018, ApJS, 237, 13, Moscoso, J., de Souza, R. S., Coc, A., & Iliadis, C. 2021, ApJ, 923,
doi: 10.3847/1538-4365/aacb24 49, doi: 10.3847/1538-4357/ac1db0
Lippuner, J., Fernández, R., Roberts, L. F., et al. 2017, MNRAS, Mossa, V., Stöckel, K., Cavanna, F., et al. 2020, Nature, 587, 210,
472, 904, doi: 10.1093/mnras/stx1987 doi: 10.1038/s41586-020-2878-4
Lippuner, J., & Roberts, L. F. 2017, ApJS, 233, 18, Mösta, P., Roberts, L. F., Halevi, G., et al. 2018, ApJ, 864, 171,
doi: 10.3847/1538-4365/aa94cb doi: 10.3847/1538-4357/aad6ec
Livne, E., & Arnett, D. 1995, ApJ, 452, 62, doi: 10.1086/176279 Mueller, E. 1986, A&A, 162, 103
Longland, R., Martin, D., & José, J. 2014, A&A, 563, A67, Mueller, E., & Arnett, W. D. 1986, ApJ, 307, 619,
doi: 10.1051/0004-6361/201321958 doi: 10.1086/164448
MacFadyen, A. I., & Woosley, S. E. 1999, ApJ, 524, 262, Müller, B. 2016, PASA, 33, e048, doi: 10.1017/pasa.2016.40
doi: 10.1086/307790 —. 2020, Living Rev. Comput. Astrophys., 6, 3,
Maeda, K., & Terada, Y. 2016, International Journal of Modern doi: 10.1007/s41115-020-0008-5
Physics D, 25, 1630024, doi: 10.1142/S021827181630024X Mumpower, M. R., Jaffke, P., Verriere, M., & Randrup, J. 2020,
Maeder, A., & Meynet, G. 2012, Rev. Mod. Phys., 84, 25, PhRvC, 101, 054607, doi: 10.1103/PhysRevC.101.054607
doi: 10.1103/RevModPhys.84.25 Mumpower, M. R., Kawano, T., Sprouse, T. M., et al. 2018, ApJ,
Mamdouh, A., Pearson, J., Rayet, M., & Tondeur, F. 2001, 679, 337 869, 14, doi: 10.3847/1538-4357/aaeaca
Myers, W. D., & Świaţecki, W. J. 1999, 60, 014606,
Marketin, T., Huther, L., & Martı́nez-Pinedo, G. 2016, PhRvC, 93,
doi: 10.1103/PhysRevC.60.014606
025805, doi: 10.1103/PhysRevC.93.025805
Nadyozhin, D. K., & Deputovich, A. Y. 2002, A&A, 386, 711,
Martin, D. 2017, PhD thesis, Technical University of Darmstadt,
doi: 10.1051/0004-6361:20011844
Darmstadt, Germany
Nagataki, S., Hashimoto, M.-a., Sato, K., & Yamada, S. 1997, ApJ,
Martin, D., Perego, A., Arcones, A., et al. 2015, ApJ, 813, 2,
486, 1026, doi: 10.1086/304565
doi: 10.1088/0004-637X/813/1/2
Nakamura, K., Takiwaki, T., Kotake, K., & Nishimura, N. 2014,
Martinez-Pinedo, G., Mocelj, D., Zinner, N., et al. 2007, Prog. Part.
ApJ, 782, 91, doi: 10.1088/0004-637X/782/2/91
Nucl. Phys., 59, 199, doi: 10.1016/j.ppnp.2007.01.018
Navó, G., Reichert, M., Obergaulinger, M., & Arcones, A. 2023,
McLaughlin, G. C., & Surman, R. 2005, NuPhA, 758, 189,
ApJ, 951, 112, doi: 10.3847/1538-4357/acd640
doi: 10.1016/j.nuclphysa.2005.05.036
Nishimura, N., Hirschi, R., Rauscher, T., St. J. Murphy, A., &
Meakin, C. A., Seitenzahl, I., Townsley, D., et al. 2009, ApJ, 693,
Cescutti, G. 2017a, MNRAS, 469, 1752,
1188, doi: 10.1088/0004-637X/693/2/1188
doi: 10.1093/mnras/stx696
Meisel, Z., George, S., Ahn, S., et al. 2020, PhRvC, 101, 052801,
Nishimura, N., Sawai, H., Takiwaki, T., Yamada, S., & Thielemann,
doi: 10.1103/PhysRevC.101.052801
F.-K. 2017b, ApJL, 836, L21, doi: 10.3847/2041-8213/aa5dee
Mendoza-Temis, J. d. J., Wu, M.-R., Langanke, K., et al. 2015,
Nishimura, N., Takiwaki, T., & Thielemann, F.-K. 2015, ApJ, 810,
PhRvC, 92, 055805, doi: 10.1103/PhysRevC.92.055805 109, doi: 10.1088/0004-637X/810/2/109
Meyer, B. S., & Adams, D. C. 2007, Meteoritics and Planetary Nishimura, S., Kotake, K., Hashimoto, M.-a., et al. 2006, ApJ, 642,
Science Supplement, 42, 5215 410, doi: 10.1086/500786
Mihalas, D. 1999, Foundations of radiation hydrodynamics, Dover Nomoto, K., Thielemann, F.-K., & Miyaji, S. 1985, A&A, 149, 239
Books on Physics (Mineola, NY: Dover Publications) Nomoto, K., Thielemann, F.-K., & Yokoi, K. 1984, ApJ, 286, 644,
Miller, J. M., Sprouse, T. M., Fryer, C. L., et al. 2020, ApJ, 902, 66, doi: 10.1086/162639
doi: 10.3847/1538-4357/abb4e3 Obergaulinger, M., & Aloy, M. Á. 2017, MNRAS, 469, L43,
Mirizzi, A., Tamborra, I., Janka, H. T., et al. 2016, Nuovo Cimento doi: 10.1093/mnrasl/slx046
Rivista Serie, 39, 1, doi: 10.1393/ncr/i2016-10120-8 —. 2021, MNRAS, 503, 4942, doi: 10.1093/mnras/stab295
Möller, P., Mumpower, M. R., Kawano, T., & Myers, W. D. 2019, Oda, T., Hino, M., Muto, K., Takahara, M., & Sato, K. 1994,
Atomic Data and Nuclear Data Tables, 125, 1, Atomic Data and Nuclear Data Tables, 56, 231,
doi: 10.1016/j.adt.2018.03.003 doi: 10.1006/adnd.1994.1007
Möller, P., Sierk, A. J., Ichikawa, T., Iwamoto, A., & Mumpower, Olive, K. A., Schramm, D. N., Steigman, G., & Walker, T. P. 1990,
M. 2015, PhRvC, 91, 024310, PhRvB, 236, 454, doi: 10.1016/0370-2693(90)90382-G
doi: 10.1103/PhysRevC.91.024310 Otsuki, K., Mathews, G. J., & Kajino, T. 2003, NewA, 8, 767,
More, J. J., Garbow, B. S., & Hillstrom, K. E. 1980, User guide for doi: 10.1016/S1384-1076(03)00065-4
MINPACK-1. [In FORTRAN], Tech. rep., Argonne National Otsuki, K., Tagoshi, H., Kajino, T., & Wanajo, S.-y. 2000, ApJ,
Lab., doi: 10.2172/6997568 533, 424, doi: 10.1086/308632
32

Pakmor, R., Kromer, M., Taubenberger, S., & Springel, V. 2013, Rauscher, T., & Thielemann, F.-K. 2000, At Data Nucl. Data
ApJL, 770, L8, doi: 10.1088/2041-8205/770/1/L8 Tables, 75, 1, doi: 10.1006/adnd.2000.0834
Panov, I. V., Freiburghaus, C., & Thielemann, F. K. 2001, NuPhA, Reichert, M. 2021, PhD thesis, Technische Universität, Darmstadt,
688, 587, doi: 10.1016/S0375-9474(01)00797-7 doi: https://doi.org/10.12921/tuprints-00014198
Panov, I. V., Kolbe, E., Pfeiffer, B., et al. 2005, Nuclear Physics A, Reichert, M., Obergaulinger, M., Aloy, M. Á., et al. 2023,
747, 633, doi: 10.1016/j.nuclphysa.2004.09.115 MNRAS, 518, 1557, doi: 10.1093/mnras/stac3185
Panov, I. V., Korneev, I. Y., Rauscher, T., et al. 2010, A&A, 513, Reichert, M., Obergaulinger, M., Eichler, M., Aloy, M. Á., &
A61, doi: 10.1051/0004-6361/200911967 Arcones, A. 2021, MNRAS, 501, 5733,
Paxton, B., Marchant, P., Schwab, J., et al. 2015, ApJS, 220, 15, doi: 10.1093/mnras/stab029
doi: 10.1088/0067-0049/220/1/15 Reichert, M., Winteler, C., Korobkin, O., et al. 2023, WinNet:
Peebles, P. J. E. 1966, ApJ, 146, 542, doi: 10.1086/148918 v1.0.1, Zenodo, doi: 10.5281/zenodo.8220549
Perego, A., Hempel, M., Fröhlich, C., et al. 2015, ApJ, 806, 275, Rembges, F., Freiburghaus, C., Rauscher, T., et al. 1997, ApJ, 484,
doi: 10.1088/0004-637X/806/2/275 412, doi: 10.1086/304300
Perego, A., Rosswog, S., Cabezón, R. M., et al. 2014, MNRAS, Romano, D., Tosi, M., Matteucci, F., & Chiappini, C. 2003,
443, 3134, doi: 10.1093/mnras/stu1352 MNRAS, 346, 295, doi: 10.1046/j.1365-2966.2003.07083.x
Petermann, I., Langanke, K., Martı́nez-Pinedo, G., et al. 2012, Röpke, F. K., & Sim, S. A. 2018, Space Sci. Rev., 214, 72,
European Physical Journal A, 48, 122, doi: 10.1007/s11214-018-0503-8
doi: 10.1140/epja/i2012-12122-6
Röpke, F. K., Kromer, M., Seitenzahl, I. R., et al. 2012, ApJL, 750,
Piessens, R., de Doncker-Kapenga, E., & Ueberhuber, C. W. 1983,
L19, doi: 10.1088/2041-8205/750/1/L19
Quadpack. A subroutine package for automatic integration
Rosswog, S. 2013, Philosophical Transactions of the Royal Society
Pignatari, M., & Hirschi, R. 2022, Weak s-process, 1.1.1, Zenodo,
of London Series A, 371, 20272, doi: 10.1098/rsta.2012.0272
doi: 10.5281/zenodo.6474728
Rosswog, S., & Korobkin, O. 2022, Annalen der Physik, n/a,
Piran, T., Nakar, E., & Rosswog, S. 2013, MNRAS, 430, 2121,
2200306, doi: https://doi.org/10.1002/andp.202200306
doi: 10.1093/mnras/stt037
Rosswog, S., Piran, T., & Nakar, E. 2013, MNRAS, 430, 2585,
Pitrou, C., Coc, A., Uzan, J.-P., & Vangioni, E. 2018, PhR, 754, 1,
doi: 10.1093/mnras/sts708
doi: 10.1016/j.physrep.2018.04.005
Ryan, S. G., Beers, T. C., Olive, K. A., Fields, B. D., & Norris, J. E.
—. 2021, MNRAS, 502, 2474, doi: 10.1093/mnras/stab135
2000, ApJL, 530, L57, doi: 10.1086/312492
Planck Collaboration, Ade, P. A. R., Aghanim, N., et al. 2016,
Sahu, B., & Bhoi, S. 2016, PhRvC, 93, 044301,
A&A, 594, A13, doi: 10.1051/0004-6361/201525830
doi: 10.1103/PhysRevC.93.044301
Plewa, T., & Müller, E. 1999, A&A, 342, 179.
Sallaska, A. L., Iliadis, C., Champange, A. E., et al. 2013, ApJS,
https://arxiv.org/abs/astro-ph/9807241
207, 18, doi: 10.1088/0067-0049/207/1/18
Potekhin, A. Y., & Chabrier, G. 2000, Phys. Rev. E, 62, 8554,
Salpeter, E. E. 1954, Australian Journal of Physics, 7, 373,
doi: 10.1103/PhysRevE.62.8554
doi: 10.1071/PH540373
Powell, J., Müller, B., Aguilera-Dena, D. R., & Langer, N. 2023,
MNRAS, 522, 6070, doi: 10.1093/mnras/stad1292 Salpeter, E. E., & van Horn, H. M. 1969, ApJ, 155, 183,
Pruet, J., & Fuller, G. M. 2003, ApJS, 149, 189, doi: 10.1086/149858
doi: 10.1086/376753 Sandoval, M. A., Hix, W. R., Messer, O. E. B., Lentz, E. J., &
Pruet, J., Woosley, S. E., & Hoffman, R. D. 2003, ApJ, 586, 1254, Harris, J. A. 2021, ApJ, 921, 113,
doi: 10.1086/367957 doi: 10.3847/1538-4357/ac1d49
Psaltis, A., Arcones, A., Montes, F., et al. 2022, ApJ, 935, 27, Schatz, H. 2022, priv. communication
doi: 10.3847/1538-4357/ac7da7 Schatz, H., Toenjes, R., Pfeiffer, B., et al. 2002, ApJ, 579, 626,
Qian, Y.-Z., & Woosley, S. E. 1996, ApJ, 471, 331, doi: 10.1086/342939
doi: 10.1086/177973 Schatz, H., Aprahamian, A., Goerres, J., et al. 1998, PhR, 294, 167,
Radice, D., Abdikamalov, E., Ott, C. D., et al. 2018, J. Phys. G, 45, doi: 10.1016/S0370-1573(97)00048-3
053003, doi: 10.1088/1361-6471/aab872 Schatz, H., Aprahamian, A., Barnard, V., et al. 2001, NuPhA, 688,
Rauscher, T. 2003, ApJS, 147, 403, doi: 10.1086/375733 150, doi: 10.1016/S0375-9474(01)00688-1
—. 2020, Essentials of Nucleosynthesis and Theoretical Nuclear Schenk, O., & Gärtner, K. 2004, Future Gener. Comput. Syst., 20,
Astrophysics, doi: 10.1088/2514-3433/ab8737 475, doi: 10.1016/j.future.2003.07.011
—. 2022, European Physical Journal A, 58, 214, Seitenzahl, I. R., Röpke, F. K., Fink, M., & Pakmor, R. 2010,
doi: 10.1140/epja/s10050-022-00866-9 MNRAS, 407, 2297, doi: 10.1111/j.1365-2966.2010.17106.x
 33

Shen, K. J., Boubert, D., Gänsicke, B. T., et al. 2018, ApJ, 865, 15, Truran, J. W., Arnett, W. D., & Cameron, A. G. W. 1967, Canadian
doi: 10.3847/1538-4357/aad55b Journal of Physics, 45, 2315, doi: 10.1139/p67-184
Siegel, D. M., Barnes, J., & Metzger, B. D. 2019, Nature, 569, 241, Truran, J. W., Cameron, A. G. W., & Gilbert, A. 1966a, Canadian
doi: 10.1038/s41586-019-1136-0 Journal of Physics, 44, 563, doi: 10.1139/p66-049
Sieverding, A., Langanke, K., Martı́nez-Pinedo, G., et al. 2019, Truran, J. W., Hansen, C. J., Cameron, A. G. W., & Gilbert, A.
ApJ, 876, 151, doi: 10.3847/1538-4357/ab17e2 1966b, Canadian Journal of Physics, 44, 151,
Sieverding, A., Martı́nez-Pinedo, G., Huther, L., Langanke, K., & doi: 10.1139/p66-011
Heger, A. 2018, ApJ, 865, 143, doi: 10.3847/1538-4357/aadd48 Vartanyan, D., Coleman, M. S. B., & Burrows, A. 2022, MNRAS,
Sieverding, A., Waldrop, P. G., Harris, J. A., et al. 2023, ApJ, 950, 510, 4689, doi: 10.1093/mnras/stab3702
34, doi: 10.3847/1538-4357/acc8d1 Vasini, A., Matteucci, F., & Spitoni, E. 2022, MNRAS, 517, 4256,
Smith, A. I., Johnson, E. T., Chen, Z., et al. 2023, ApJ, 947, 65, doi: 10.1093/mnras/stac2981
doi: 10.3847/1538-4357/acbaff Vassh, N., Vogt, R., Surman, R., et al. 2019, Journal of Physics G
Smith, M. S., Kawano, L. H., & Malaney, R. A. 1993, ApJS, 85, Nuclear Physics, 46, 065202, doi: 10.1088/1361-6471/ab0bea
219, doi: 10.1086/191763 Viola, V., & Seaborg, G. 1966, Journal of Inorganic and Nuclear
Sobiczewski, A., Patyk, Z., & Ćwiok, S. 1989, Physics Letters B, Chemistry, 28, 741,
224, 1, doi: 10.1016/0370-2693(89)91038-1 doi: https://doi.org/10.1016/0022-1902(66)80412-8
Sprouse, T. M., Mumpower, M. R., & Surman, R. 2021, PhRvC,
Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature
104, 015803, doi: 10.1103/PhysRevC.104.015803
Methods, 17, 261, doi: 10.1038/s41592-019-0686-2
Sukhbold, T., Ertl, T., Woosley, S. E., Brown, J. M., & Janka, H.-T.
Vonlanthen, P., Rauscher, T., Winteler, C., et al. 2009, A&A, 503,
2016, ApJ, 821, 38, doi: 10.3847/0004-637X/821/1/38
47, doi: 10.1051/0004-6361/200811297
Surman, R., & McLaughlin, G. C. 2004, ApJ, 603, 611,
Wagoner, R. V., Fowler, W. A., & Hoyle, F. 1967, ApJ, 148, 3,
doi: 10.1086/381672
doi: 10.1086/149126
Suzuki, T., Toki, H., & Nomoto, K. 2016, ApJ, 817, 163,
Walker, T. P., Steigman, G., Schramm, D. N., Olive, K. A., &
doi: 10.3847/0004-637X/817/2/163
Kang, H.-S. 1991, ApJ, 376, 51, doi: 10.1086/170255
Tamborra, I., Müller, B., Hüdepohl, L., Janka, H.-T., & Raffelt, G.
Wanajo, S., Hirai, Y., & Prantzos, N. 2021, MNRAS, 505, 5862,
2012, PhRvD, 86, 125031, doi: 10.1103/PhysRevD.86.125031
doi: 10.1093/mnras/stab1655
Thielemann, F. K. 1980, PhD thesis, -
Wanajo, S., Kajino, T., Mathews, G. J., & Otsuki, K. 2001, ApJ,
Thielemann, F. K., Arnould, M., & Hillebrandt, W. 1979, A&A,
554, 578, doi: 10.1086/321339
74, 175
Wanajo, S., Müller, B., Janka, H.-T., & Heger, A. 2018, ApJ, 852,
Thielemann, F.-K., Diehl, R., Heger, A., Hirschi, R., &
40, doi: 10.3847/1538-4357/aa9d97
Liebendoerfer, M. 2018a, in Astrophysics and Space Science
Library, Vol. 453, Astrophysics with Radioactive Isotopes, ed. Wiescher, M., Gorres, J., Thielemann, F.-K., & Ritter, H. 1986,
R. Diehl, D. H. Hartmann, & N. Prantzos (Cham: Springer), A&A, 160, 56
173–286, doi: 10.1007/978-3-319-91929-4 4 Winteler, C. 2012, PhD thesis, The University of Basel, Basel,
Thielemann, F.-K., Isern, J., Perego, A., & von Ballmoos, P. 2018b, Switzerland
Space Sci. Rev., 214, 62, doi: 10.1007/s11214-018-0494-5 Winteler, C., Käppeli, R., Perego, A., et al. 2012, ApJL, 750, L22,
Thielemann, F.-K., Nomoto, K., & Hashimoto, M.-A. 1996, ApJ, doi: 10.1088/2041-8205/750/1/L22
460, 408, doi: 10.1086/176980 Witt, M., Psaltis, A., Yasin, H., et al. 2021, ApJ, 921, 19,
Thielemann, F.-K., Nomoto, K., & Yokoi, K. 1986, A&A, 158, 17 doi: 10.3847/1538-4357/ac1a6d
Thompson, T. A., Burrows, A., & Meyer, B. S. 2001, ApJ, 562, Woosley, S. E., Arnett, W. D., & Clayton, D. D. 1973, ApJS, 26,
887, doi: 10.1086/323861 231, doi: 10.1086/190282
Tiesinga, E., Mohr, P. J., Newell, D. B., & Taylor, B. N. 2021, Woosley, S. E., & Heger, A. 2006, ApJ, 637, 914,
Reviews of Modern Physics, 93, 025010, doi: 10.1086/498500
doi: 10.1103/RevModPhys.93.025010 Woosley, S. E., Heger, A., & Weaver, T. A. 2002, Reviews of
Timmes, F. X. 1999, ApJS, 124, 241, doi: 10.1086/313257 Modern Physics, 74, 1015, doi: 10.1103/RevModPhys.74.1015
Timmes, F. X., & Arnett, D. 1999, ApJS, 125, 277, Woosley, S. E., Taam, R. E., & Weaver, T. A. 1986, ApJ, 301, 601,
doi: 10.1086/313271 doi: 10.1086/163926
Timmes, F. X., Hoffman, R. D., & Woosley, S. E. 2000, ApJS, 129, Woosley, S. E., & Weaver, T. A. 1995, ApJS, 101, 181,
377, doi: 10.1086/313407 doi: 10.1086/192237
34

Wu, M.-R., Barnes, J., Martı́nez-Pinedo, G., & Metzger, B. D. Yakovlev, D. G., & Shalybkov, D. A. 1989,
2019, PhRvL, 122, 062701, Astrophys. Space Phys. Res., 7, 311
doi: 10.1103/PhysRevLett.122.062701
Wu, M.-R., Fernández, R., Martı́nez-Pinedo, G., & Metzger, B. D. Yang, J., Turner, M. S., Steigman, G., Schramm, D. N., & Olive,
2016, MNRAS, 463, 2323, doi: 10.1093/mnras/stw2156
K. A. 1984, ApJ, 281, 493, doi: 10.1086/162123
Xu, Y., Takahashi, K., Goriely, S., et al. 2013, NuPhA, 918, 61,
doi: 10.1016/j.nuclphysa.2013.09.007 Zenati, Y., Siegel, D. M., Metzger, B. D., & Perets, H. B. 2020,
Yakovlev, D. G., Gasques, L. R., Afanasjev, A. V., Beard, M., &
MNRAS, 499, 4097, doi: 10.1093/mnras/staa3002
Wiescher, M. 2006, Phys. Rev. C, 74, 035803,
doi: 10.1103/PhysRevC.74.035803
 35

10 8 10 6 10 4 10 2 10 8 10 6 10 4 10 2
10 7 10 5 10 3 10 1 10 7 10 5 10 3 10 1

10 2 10 2
Mass fraction

Mass fraction
10 5 10 5

10 8 10 8

101 101
10 1 10 1
0 0
104

104
10 11 10 11
10 10
0 50 100 150 200 0 50 100 150 200
Mass number Mass number

Figure 23. Left plot: calculation using the implicit Euler integration scheme using the Newton-Raphson convergence criteria that is based
on baryon conservation (Eq. (54)) and different values of 𝜖 NR . Right plot: the same, but using an alternative convergence criteria of the
𝑋1
Newton-Raphson | max(® 𝑥 𝑘+1 , 𝑥®𝑘 )/min(®
𝑥 𝑘+1 , 𝑥®𝑘+1 ) − 1| < 𝜖NR . The lower panels show the deviation defined as Δ = 1 − 𝑋2
, using the most
restrictive parameters as reference (𝑋1 ).

APPENDIX

A. CODE CONVERGENCE
The accuracy of the nucleosynthesis results not only depends on the nuclear input, but also on numerical parameters. In
the following, we investigate in more detail the latter error. For this we use a neutron-rich trajectory from an MR-SNe of the
simulations of Winteler et al. (2012). We use reactions from the Jina Reaclib (Cyburt et al. 2010) with additional 𝛼-decays from
the Viola-Seaborg formula, theoretical weak rates from Langanke & Martı́nez-Pinedo (2001) that we exchange with experimental
reaction rates at 10−1 GK. Fission rates have been used as described in Section 4.2.7 with the fragment distribution of Panov et al.
(2001).
First we will investigate different values of 𝜖 NR for the convergence criterion of the root-finding algorithm within the implicit
Euler method (Eq. (54)). As default in WinNet, we perform at least two root-finding iterations. To avoid a re-adjustment to
smaller and smaller time steps due to a not converged root-finding (see Fig. 3), we set the maximum amount of allowed root-finding
iterations to a large value of 1000.
The final mass fractions of all runs are shown in the upper-left panel of Fig. 23, the difference is defined by Δ = 1 − 𝑋 1
𝑋2 , where
−8
we took 𝑋1 as mass fractions from the run with 𝜖NR = 10 . This is shown in the lower-left panel of Fig. 23. The maximum
deviation is of the order of ∼ 0.1%. Interestingly, it is the iron region that is prone to errors. This part of enhanced errors vanishes
completely when not using theoretical weak rates. These rates depend on temperatures as well as densities and can therefore be
more challenging to integrate. For values 𝜖NR > 10−7 , there is no difference visible. This is due to the fact that the mass for these
precisions is already conserved within the minimum of two Newton-Raphson iterations.
An alternative convergence criteria of the Newton-Raphson that is not based on baryon conservation is
| max( 𝑥®𝑘+1 , 𝑥®𝑘 )/min( 𝑥®𝑘+1 , 𝑥®𝑘+1 ) − 1| < 𝜖 NR for 𝑥®𝑘+1 > 10−10 . In other words, every abundance should be converged within
a given percentage. The result for this convergence criteria is shown in the right panels of Fig. 23. Again, the difference between
the most restrictive case and the least restrictive one is of the order of ∼ 0.1%, but the parameter has a much more direct impact
on the accuracy. The most restrictive scenarios of both convergence criteria agree even within ∼ 0.01% which demonstrates that
both criteria can be used interchangeably, and we therefore only include the criterion that is based on baryon conservation, as it
has a better performance.
All previous calculations were done with the same time step factor of 𝜖Euler = 0.1 (Eq. (57)). We reduced this factor and tested
values of 5 × 10−2 , 1 × 10−2 , and 5 × 10−3 . As shown in the left panels of Fig. 24, the abundances are converged within ∼ 10%.
36

5 10 3 10 2 5 10 2 10 9 10 7 10 5 10 3
10 1
10 8 10 6 10 4 10 2

10 2 10 2

Mass fraction
Mass fraction

10 5 10 5

10 8 10 8

101 101
10 1 10 1
0 0
10 11

104
10 11
104

10 10
0 50 100 150 200 0 50 100 150 200
Mass number Mass number

Figure 24. Left plot: calculation using the implicit Euler integration scheme using different time steps by varying 𝜖 Euler (Eq. (57)). Right plot:
𝑋1
the same, but using the Gear solver with different time steps by varying 𝜖Gear . The lower panels show the deviation defined as Δ = 1 − 𝑋 2
, using
the most restrictive parameters as reference (𝑋1 ).

Gear Euler

10 2
Mass fraction

10 5

10 8

1041
10
0
104

1014
10
0 50 100 150 200
Mass number

Figure 25. Comparison of a calculation using the implicit Euler method with 𝜖 Euler = 5 × 10 −3 and a calculation using Gear’s method with
𝜖 Gear = 10 −9 . We note that the Gear solver in the lower panel is a horizontal line by definition.

In practice, it is not feasible to use a factor of 𝜖Euler = 5 × 10−3 when calculating many trajectories (c.f., ∼ 3000 versus ∼ 60000
time steps for 𝜖 Euler = 10−1 and 𝜖 Euler = 5 × 10−3 , respectively).
The Gear integration method, on the other hand, estimates the time step in a more sophisticated way, based on integration
errors. This error is controlled by 𝜖Gear (Eq. (70)). When reducing 𝜖Gear , one directly controls the numerical error (right panels
of Fig. 24). The error can be reduced to an almost arbitrary precision, and for all calculated runs it lies within an astonishing
precision of ∼ 0.1%.
Finally, it is interesting to compare the most precise calculation using the Gear solver (𝜖Gear = 10−9 ) with the most precise
calculation using the implicit Euler method (𝜖 Euler = 5 × 10−3 ). This comparison is shown in Fig. 25. The difference between the
calculation using the Gear solver and the implicit Euler is for most parts within 10%; however some regions, i.e., around 𝐴 ∼ 70
and 𝐴 ∼ 140, are differing by a factor of ∼ 2. However, the largest deviation is visible in the abundance of protons with a factor
of 30 showing that the numerical method can also have a strong impact especially on light nuclei such as neutrons, protons, and
alphas.