How many nucleosynthesis processes exist at low metallicity?
aa r X i v : . [ a s t r o - ph . S R ] J a n Draft version August 26, 2018
Preprint typeset using L A TEX style emulateapj v. 5/2/11
HOW MANY NUCLEOSYNTHESIS PROCESSES EXIST AT LOW METALLICITY?
C. J. Hansen
Landessternwarte, ZAH, Heidelberg University, K¨onigstuhl 12, 69117 Heidelberg, GermanyDark Cosmology Centre, The Niels Bohr Institute, Copenhagen, Denmark
F. Montes
Joint Institute for Nuclear Astrophysics, Michigan State University, East Lansing, MI 48824, USANational Superconducting Cyclotron Laboratory, Michigan State University, East Lansing, MI 48824, USA
A. Arcones
Institut f¨ur Kernphysik, Technische Universit¨at Darmstadt, Schlossgartenstr. 2, Darmstadt 64289, GermanyGSI Helmholtzzentrum f¨ur Schwerionenforschung GmbH, Planckstr. 1, Darmstadt 64291, Germany (Dated: today)
Draft version August 26, 2018
ABSTRACTAbundances of low-metallicity stars offer a unique opportunity to understand the contribution andconditions of the different processes that synthesize heavy elements. Many old, metal-poor stars showa robust abundance pattern for elements heavier than Ba, and a less robust pattern between Sr andAg. Here we probe if two nucleosynthesis processes are sufficient to explain the stellar abundancesat low metallicity, and we carry out a site independent approach to separate the contribution fromthese two processes or components to the total observationally derived abundances. Our approachprovides a method to determine the contribution of each process to the production of elements suchas Sr, Zr, Ba, and Eu. We explore the observed star-to-star abundance scatter as a function ofmetallicity that each process leads to. Moreover, we use the deduced abundance pattern of one of thenucleosynthesis components to constrain the astrophysical conditions of neutrino-driven winds fromcore-collapse supernovae.
Subject headings:
Galaxy: evolution Galaxy: stellar content nuclear reactions, nucleosynthesis,abundances stars: abundances supernovae: general INTRODUCTION
Observations of stellar abundances at low metallici-ties are necessary to understand the origin of elementsheavier than iron (Spite & Spite 1978; Truran 1981;Ryan et al. 1996; Burris et al. 2000; Fulbright 2002;Honda et al. 2004; Barklem et al. 2005; Fran¸cois et al.2007; Lai et al. 2008; Hansen et al. 2012; Yong et al.2013; Roederer et al. 2014). At early times, those ele-ments originated from one or more primary processes,which imply that no seed nuclei needs to be producedprior to the nucleosynthesis process because the processcan create the seeds itself. An example of this is the r pro-cess; seed nuclei are synthesized starting with neutronsand protons by charged-particle reactions combined withneutron captures. Note that the s process is also observedat low metallicities even if it is a secondary process, how-ever, such observations are mostly related to binary sys-tems (see, e.g., Beers & Christlieb 2005; Masseron et al.2010; Bisterzo et al. 2011 for a discussion).The r process seems to be robust based on observationsof old r-II stars , which show an enrichment in heavyelements ( Z > [email protected], [email protected]@[email protected] see Beers & Christlieb (2005) for a definition are several indications that this robustness cannot begeneralized to all elements. Independent indications ofseveral different processes were initially found in me-teorites (Wasserburg et al. 1996). Here, we highlighttwo additional indications. First, the abundance pat-tern between Sr and Ag varies among stars, even if thosehave a very uniform pattern for heavy elements (e.g.,Sneden et al. 2008; Hansen et al. 2012, 2014). Second,improved observations have demonstrated that the abun-dances for the heaviest elements ( Z >
50) can vary signif-icantly compared to the 38 < Z <
50 elements, as shownin Fig. 1 (see also Aoki et al. (2005); Roederer et al.(2010)). This lack of robustness has raised the question:how many primary processes contribute to the abun-dances observed at low metallicities?Here, we provide a possibility to explain the variety instellar abundances at low metallicities and trace individ-ual processes, an option that only isotopic abundancesotherwise offer. Our study focuses on the abundancepatterns from a large sample of metal-poor stars. Wefollow the nomenclature of Qian & Wasserburg (2001,2007) using an H- and L-component to explain the for-mation of the heavy (
Z >
50) and lighter heavy (38 50) elements, respectively. We adopt a simple ap-proach to explain the observationally derived abundancesin metal-poor stars; only two nucleosynthesis contribu-tions (the H- and L-components) are needed to reproducethe stellar abundances within an estimated uncertainty( ± . 32 dex). In practice, we use a linear superposition ofthe H- and L-component (see also Li et al. (2013)) to ex-plain the stellar abundance pattern from the sample com-piled by Frebel et al. (2010) (after applying five selectioncriteria to remove contamination from s processes, self-pollution due to stellar mixing processes, etc.). Theseassumptions, though simple, are sufficient to explain thestellar abundance patterns from observations for most ofthe metal-poor stars passing the selection criteria.There are several nucleosynthesis pro-cesses/astrophysical sites that are H- and L-componentcandidates. The H component is most likely the r processthat produces heavy elements up to U via rapid neutroncaptures compared to beta decays. Although thisprocess was already proposed in 1957 (Burbidge et al.1957), there are still open questions concerning theastrophysical site and the neutron-rich nuclei involved(Arnould et al. 2007). The best studied site (afterthe work of Woosley et al. (1994)) is core-collapsesupernovae and their neutrino-driven winds. However,the conditions reached in these environments are notsufficiently neutron rich to produce heavy elementsup to uranium (see Arcones & Thielemann 2013, andreferences therein). Neutrino-driven winds may beslightly neutron rich or even proton rich (Roberts et al.2012; Mart´ınez-Pinedo et al. 2012). Another possibil-ity to produce the heaviest elements in core-collapsesupernovae are explosions driven by magnetic fields(see, e.g., Winteler et al. 2012). Mergers of two neutronstars or a neutron star and a black hole are alsopromising candidates to produce heavy r-process ele-ments (Lattimer & Schramm 1974; Freiburghaus et al.1999; Korobkin et al. 2012; Bauswein et al. 2013;Hotokezaka et al. 2013).While the H component is associated with the r pro-cess, the L component may be one of several pro-cesses. One possible site to produce the L compo-nent is neutrino-driven wind in core collapse super-novae. This possibility is explored in this paper. Inthese events, the alpha process (Woosley & Hoffman1992; Witti et al. 1994; Hoffman et al. 1996) or charged-particle reactions (CPR; Qian & Wasserburg 2007) pro-duce seed nuclei. Later, if the conditions are slightlyneutron rich, a weak r process can form elements upto Z ∼ 50 (see Farouqi et al. (2010); Arcones & Bliss(2014)). In proton-rich conditions, the νp process canalso reach those nuclei (Fr¨ohlich et al. 2006; Pruet et al.2006; Wanajo 2006). In both cases, it is possible toproduce an abundance pattern similar to the L com-ponent (Arcones & Montes 2011; Wanajo et al. 2011).Therefore, the L component may be the weak r pro-cess or νp process, or even a combination of both.Note that these processes may also occur in neutronstar mergers Perego et al. (2014); Just et al. (2014);Metzger & Fern´andez (2014).The L component could also be what Travaglio et al.(2004) called LEPP for Lighter Element Primary Pro-cess, and it was used to explain the contribution ofa process to the abundances from Sr to Ag (see alsoMontes et al. (2007) and Hansen et al. (2012)). Al-though the initial motivation of the LEPP was toexplain missing solar abundances using stellar mod-els, it has been pointed out by, e.g., Trippella et al.(2014) that such a process may not be needed af-ter all. Nevertheless, additional possibilities to pro- -3 -2 -1 a b un d a n c e b a 35 40 45 50 55 60 65 70Z a b un d a n c e H-component L - c o m p o n e n t Fig. 1.— Two metal-poor stellar abundance patterns are shownin the upper panel compared to the solar r process (grey thin line).The pattern “a” (red line and squares) corresponds to CS 22892-052, and the “b” pattern (blue line and circles) to HD 122563.All abundances are normalized to Sr. The bottom panel showsa sketch of the components that may contribute to produce thesetwo patterns. The upper one is the H component (red line) and thelower one the L component (blue). The solid lines indicate thatelements in this range are mainly produced by this component,while the dashed lines indicate that the component may not paythe predominant contribution to this elemental range. duce an L component may be a primary s pro-cess in fast rotating stars (Frischknecht et al. 2012;Pignatari et al. 2008), or an early s process leadingto a mass transfer in an extremely metal-poor bi-nary star system (Straniero et al. 2004; Lucatello et al.2005; Masseron et al. 2010; Bisterzo et al. 2010, 2011;Stancliffe et al. 2011; Cruz et al. 2013).In this paper, we extract the L and H components fromthe metal-poor stellar abundance patterns as indicated inFig. 1. As such, the extracted abundance pattern is site-independent. Although we explore neutrino-driven windsas an L component possibility, we do not exclude otherprocesses/sites from being suitable candidates. The pat-tern “a” (red line and squares) corresponds to CS 22892-052 (Sneden et al. 2003) with an enhancement of heavyelements and a robust pattern compared to the solarscaled pattern. The “b” pattern (blue line and cir-cles) corresponds to HD 122563 (Wallerstein et al. 1963;Honda et al. 2004, 2007) and is characterized by higherabundances of the lighter heavy elements ( Z < 50) andmuch lower abundances for Z > DATA—OBSERVATIONALLY DERIVED STELLARABUNDANCES To minimize the number of contributing processes, weonly consider metal-poor (old) stars that are less enrichedby s processes compared to younger (more metal-rich)stars. To clean the sample presented in Frebel et al.(2010), we apply five selection criteria (below) to thelarge, inhomogeneous sample:1. [Fe/H] < − . 5: this cut removes the majority ofthe s-process contribution. Travaglio et al. (2004)state that the s process(es) contributes little to halostars below [Fe/H]= − . 5. Recently, Hansen et al.(2014) found that the s process yields may be trace-able below [Fe/H]= − . − . − . − . − 3) ensuresthat a large fraction of heavy elements are still de-tectable, even if the star is not extremely enhancedin neutron-capture elements. Most absorption linesweaken as metallicity decreases, making the abun-dance analysis increasingly difficult with decreasingmetallicity.2. [C/Fe] < . 7: this ensures that no carbon enhancedmetal-poor (CEMP) star is included (Aoki et al.2007), as most Fe-poor stars are carbon-enhancedstars, many of which are also s-process enriched.3. [Ba/Fe] < . 0: this cuts out Ba stars and togetherwith the previous criteria removes strong s-processenhancement in, e.g., CEMP-s stars.4. Excluding abundances that are only upper lim-its yields a better and more solid final abundancepattern with known reasonable sized uncertainties.This facilitates a more direct comparison of obser-vations and predictions.5. Most heavy elements are detected in the extendedatmospheres of giant stars, that due to stellar evo-lution have had their surface composition altered.This change is normally seen in their carbon andnitrogen abundances, which is why we place thecuts: [C/N] < − . > . 5. These ex-clude stars with internal mixing owing to the stel-lar evolution (Spite et al. 2005). Very evolved stars(giants) burn C into N and later O, which will re-sult in lower C and higher N and O abundances.The original sample from Frebel et al. (2010) is a com-pilation of different sources from the literature. There-fore, our reduced sample is inhomogeneous owing to thevariety of different stellar parameter scales and methodsused to derive these abundances. After carefully exam-ining the observational data, inconsistencies between the “original” data found in the literature and the compila-tion in Frebel et al. (2010) were revealed for two of ourreduced sample’s stars. As a consequence, these were re-moved from our final sample. The star CS 30325-094 hasa Eu abundance that is observed only as an upper limitin Fran¸cois et al. (2007) and the Pm abundance was notfound in the quoted reference. In addition, CS 22783-055 (McWilliam et al. 1995b,a) has abundances in theFrebel et al. 2010 table that we were not able to find inthe literature.If we also require that each star needs to have at leastfive heavy element detections or more (i.e., we do notcount upper limits), the final reduced sample consists of39 stars. However, if we loosen criteria (5) to only affect[C/N] < − . 4, and set no [N/Fe] constraint, the sampleis increased to 53 stars (model 4 in Table 1).The final abundance pattern of each star, consistsof both neutral and ionized elemental abundances, andsome species (e.g., Sr I—the minority of Sr) are moreaffected by nonlocal thermodynamic equilibrium (NLTEand possibly three-dimensional (3D)) effects than otherspecies (e.g., Sr II—the majority of Sr). This introducesa possible bias between the mixture of neutral/ionizedelements that compose the total abundance pattern, andthis bias may exceed the uncertainties stemming fromthe inhomogeneity of the sample. The NLTE correc-tions are not calculated for all the heavy elements (ow-ing to the lack of atomic physics), and even fewer ofthe heavy elements have NLTE corrections calculatedfor a large stellar parameter space (some of these ele-ments are Sr and Ba which have been investigated in de-tail, e.g., Andrievsky et al. 2009, 2011; Bergemann et al.2012; Hansen et al. 2013). Fortunately, several of theheavy elements have abundances derived from the ma-jority species (which, in many cases, are single ionizedlines), and for some of these elements the 3D and NLTEeffects may cancel out, thereby removing the bias in theheavy element abundance pattern.We account for the sample inhomogeneity that maylead to biases in the abundance pattern by propagatingslightly increased uncertainties into the nucleosynthesiscomponents in the next sections (Sect. 3 and AppendixA). NUCLEOSYNTHESIS COMPONENTS Following Qian & Wasserburg (2001, 2007, 2008), wecall the two components that contribute to the metal-poor stars the L component and H component (see Fig. 1for schematic representation). We assume that the Hcomponent is the main source of heavy r process ele-ments ( Z > < Z < Z > 50 elements.With these two components, there are several possibili-ties to explain the typical patterns shown in the upperpanel of Fig. 1. The pattern “a” can be produced only byan H component going from Sr to the heaviest elements.Another possibility to explain the “a” pattern would bea combination of an H component that contributes only The log ǫ abundances were adopted and when necessary con-verted to relative abundances using the solar abundances fromAnders & Grevesse (1989). to the heavy part ( Z > 50) and a L component to ex-plain the pattern below Z < 50. The “b” pattern can bethe result of a single L component that extends towardheavy elements or the combination of an L componentup to Z = 50 and a small contribution from the H com-ponent to explain the low abundances of heavy elements. Component identification In order to explain the abundances of our reduced sam-ple of stars described in Sect. 2 by an L and H compo-nent, we first extract the abundance pattern of these twocomponents. The pure nucleosynthesis component pat-terns are obtained using three different methods (M1,M2, M3) in order to test their robustness and to esti-mate their uncertainties (see Appendix A for details).All methods use abundances from the metal-poor starsHD 122563 and HD 88609 (which have large [Sr/Eu] ra-tios, Honda et al. 2007) and CS 22892-052 (which has alarge [Eu/Fe] ratio, Sneden et al. 2003).Method one (M1) assumes that HD 122563 has onlybeen enriched by the L component (due to the large Sr-enrichment) while CS 22892-052 has only been enrichedby the H component (due to its large Eu-enrichment). Assuch, their abundances already show the individual com-ponents. These are shown in the upper panel of Fig. 2.Method two (M2) follows Montes et al. (2007) and as-sumes that while CS 22892-052 has a pure H-componentabundance, HD 122563 shows a large L component com-bined with a small H-component contribution. Thissmall contribution is removed by subtracting the abun-dances of CS 22892-052 from the HD 122563 abundance(by scaling to the average of Eu, Gd, Dy, Er, and Ybabundances in each star). This leaves only the pureL-component pattern as shown in the middle panel ofFig. 2.Method three (M3) follows Li et al. (2013) and as-sumes that the mentioned metal-poor stars’ abundancesdo not have pure nucleosynthesis components, but in-stead have a dominant contribution from one of the com-ponents. The pure component abundances are obtainedby systematically eliminating the abundances from theprocess that contributes the least. The first-order L-component abundance is obtained by subtracting theCS 22892-052 abundances (scaled to the Eu abundancewhich is predominantly produced by the H component)from the average of the HD 122563 and HD 88609 abun-dances. Conversely, the first-order H-component abun-dance is obtained by subtracting the average of theHD 122563 and HD 88609 abundances (scaled to the Feabundance) from the CS 22892-052 abundances. Theindividual components are obtained by further subtrac-tion of the remaining contribution, e.g., the n th -orderL-component abundance is obtained by subtracting the( n − th -order H-component abundances scaled to Eufrom the average of the HD 122563 and HD 88609 stellarabundances. The procedure is repeated until the differ-ences in both the L component and H component aresmaller than the observational error. Figure 2 (bottompanel) shows the nucleosynthesis components obtainedfollowing this method. The robustness of the derivedcomponents using method three was checked by using the same applies for HD 88609 −3.0−2.5−2.0−1.5−1.0−0.50.0 l o g ǫ M1 L componentH component −3.0−2.5−2.0−1.5−1.0−0.50.0 l o g ǫ M235 40 45 50 55 60 65 70Z−3.0−2.5−2.0−1.5−1.0−0.50.0 l o g ǫ M3 Fig. 2.— Pure L component and H component obtained usingthe three methods described in the text. In all three panels, thesolid blue (dashed green) line corresponds to the L(H) componentobtained in method 1 (M1), while the symbols vary with the meth-ods. different combinations of metal-poor stars. Since the as-sumption is that all stellar abundances at low metallicityhave contributions from robust L and H components, anypair of metal-poor stars could, in principle, be used toobtain the pure components (see Appendix A for addi-tional tests). However, the errors of the iterative methodare smaller when the differences between the observedabundances are larger.The derived H-component abundances shown in Fig 2are remarkably consistent between the different methods.The calculated abundance difference between methods iswithin ± . ± . ± . ∼ . < 50. However, we refer to Sect. 4.1 andFig. 10 for further discussion on the robustness of the Lcomponent. We stress that the robustness of the L com-ponent is an assumption that we will test in the followingsections. Fig. 3.— Abundances normalized to Sr for four well-known L-component stars. Fitting observations with two components The H and L components introduced in the previoussection are assumed to be responsible for the abundances( Y ) of metal-poor stars. For every element with Z > Y H and Y L ): Y calc ( Z ) = (cid:0) C H Y H ( Z ) + C L Y L ( Z ) (cid:1) × [Fe / H] , (1)where C H and C L are the weights of the H and L com-ponents to the abundances of the star, respectively. Itshould be noted that since there is an arbitrary scalingfactor when defining Y H ( Z ) and Y L ( Z ), the values of C H and C L are relative and only their overall trends havephysical significance. The factor 10 [Fe / H] is introducedto normalize the abundances at different metallicities.In order to find the coefficients C H and C L that bestmatch the observationally derived abundances in metal-poor stars, the following χ -distribution was minimized, χ = 1 ν X Z range (cid:0) log Y observed ( Z ) − log Y calc ( Z ) (cid:1) / ∆( Z ) , (2)where Z range is the elemental range considered in theminimization, ∆( Z ) corresponds to the abundance un-certainty of element Z from both the observation andthe nucleosynthesis component determination, and ν isthe number of degrees of freedom in the fit (numberof elements observed in Z range minus the number of fit-ted coefficients C ). The uncertainty in the observation(0.25 dex) and the intrinsic error in the component es-timation (0.2 dex) were added in quadrature to obtain∆(Z) = 0.32 dex for all elements, see Appendix A. Once the χ -distribution (Eq. (2)) has been minimizedfor a given star, the minimum χ obtained with the pre-ferred C H and C L coefficients is associated with thatstar. A star that has its abundances well (badly) calcu-lated within this approach, would result in a low (high) χ value. A large number of stars with high χ valueswould indicate that the assumptions in our approach areincorrect. BS16089-013= 0.15 χ -3-2-10 l og ∈ l og ∈ -3-2-101 L-componentH-componentFit χ = 3.98 35 7570656055504540 Atomic number Fig. 4.— Poor (top) and good (bottom) χ fit to two samplestars. We calculate the χ -distribution (Eq. (2)) for the starsin our sample. We have studied various models that as-sume different elemental ranges for every component anduse M3 for the H component and vary the method for theL component. These models are summarized in Table 1.In model 4, the number of stars in the sample is largerand marked with an asterisk to indicate that stars inthis model fulfill criteria 1 to 4, but not 5 (the criteriaare described in Sect. 2).Stars with low χ values show a typical trend (an exam-ple of this is shown in the lower panel of Fig. 4): Z ≥ ≤ Z ≤ Z > 50. In addition, in the range 38 ≤ Z ≤ 47, theH- and L-component abundances are remarkably similarand within the error bar ∆(Z) used in the fit. Since theL component is not making significant contributions toZ > 56 (as can be seen from the good fits using models1 and 2) a combination of L- and H-component contri-butions in the range of 38 ≤ Z ≤ 47, is almost equivalentto having a single process making only those abundances(Z ≤ 47) independently of the process responsible for theZ ≥ 56 abundances (model 3 in Table 1). TABLE 1Details (component, sample size, and χ ) of the fourmodels. H- L- Num. of Stars withmodel component component stars in sample χ ≥ χ < < > 47 M3, Z < 56 39 stars < ∗ In order to test how well our models fit the stellar abun-dances, the expected χ -probability distribution was cal-culated by adding up the expected χ -distribution ofevery star considered. Each star has an expected χ -distribution that depends only on the degrees of free-dom. The expected χ -probability distribution can thusbe expressed as f ( χ ) = X i ν i ν i / Γ( ν i / e − χ ν i / ( χ ν i ) ( ν i / − , (3)where ν i is the number of degrees of freedom for star i .The χ test relies on the assumption that each elementalabundance is normally distributed within a given error.If the expected χ -probability distribution is too largecompared to the minimum χ values satisfying Eq. (2),we either conclude that a statistically improbable excur-sion of χ has occurred, or that our model is incorrect.If, on the other hand, the expected χ -probability distri-bution is too small, it is not indicative of a poor model,but that a statistically improbable excursion of χ hasoccurred, or that ∆(Z) has been overestimated in themodel.The minimum χ is shown as a function of metallicityin the right panel of Fig. 5 for the first three models listedin Table 1. In the left panel, one can see the expected χ -distribution using Eq. (3) (red line) and histogramsof the χ -distribution values based on models 1, 2, and3 in Table 1. The stellar sample contains only stars withat least five elemental abundance detections in the rangeZ ≥ 38. This guaranteed that the fit had at least threedegrees of freedom after C H and C L were fitted. The to-tal number of stars satisfying these restrictions combinedwith the five criteria outlined in Sect. 2 is 39 stars. 10 1Counts [Fe/H]0.1110 c - distribution-3.4 -2.4-2.6-2.8-3-3.2 Fig. 5.— Minimum χ values as a function of metallicity areshown in the right panel. Left panel shows expected χ -distributionusing Eq. (3) (red line) and histograms of the χ -distribution valuesfor different models presented in Table 1. Model 1 is shown as asolid line (left panel) and black solid circles (right panel), model 2as a dashed line (left panel) and red open circles (right panel), andmodel 3 as a dotted line (left panel) and black open squares (rightpanel). In order to test if the model is valid, we comparethe number of stars with a χ value outside the range,where we expect to find 95% of the stars (stars with χ ≥ χ = 2 . χ - probability distribution (see Fig. 5 and Table 1). Only1 star, CS 22189-009, is outside the expected range of χ values. Therefore, this star cannot be explained byour assumption of two robust components. We have alsotested an increment in the minimum number of elementsobserved. When this number is increased to ≥ χ values.The good agreement gives credence to the assumption oftwo independent robust processes being responsible forZ ≥ 38 abundances in most but not all stars.The elemental range of the two components is testedin model 3 (Table 1). Here we obtain good fits whenusing the M3-H-component only for Z > 47 and the M3-L-component only for Z < 56 abundances. Without anyoverlap in the components a good fit is still obtained for38 out of the 39 stars considered (see dotted line in theleft panel of Fig. 5 and black open squares in the rightpanel). Therefore, using our method it is not possible toconstrain the elemental range of the components.We note, that if the criteria of excluding stars with in-ternal mixing (Sect. 2) is removed (model 4, Table 1), thenumber of stars with χ values larger than χ ≥ . 31 in-creases to six stars: CS 29518-051 (see Fig. 4), CS 22189-009, CS 22169-035, CS 30325-094, CS 30322-023, andCS 22783-055. Hence, internal mixing of the giant starswill show a clear effect in the overall abundance pattern,which is detectable using this method.Generally, we obtained good fits to the observationsassuming two dominant, robust components, but we canneither prove nor refute the presence of more processesnor can we at this state demonstrate the robustness ofthe processes. The good χ indicates that two processescould be sufficient and we discuss further in Sect. 4.1 thepossible number of primary processes and their robust-ness. RESULTS AND APPLICATIONS OF THE METHOD The star-to-star scatter It is well known from earlier studies (see e.g.,Spite & Spite 1978; Truran 1981; McWilliam et al.1995b; Ryan et al. 1996; Fulbright 2002; Barklem et al.2005; Fran¸cois et al. 2007; Roederer et al. 2010;Hansen et al. 2012; Yong et al. 2013; Roederer et al.2014) that a large star-to-star abundance scatter as afunction of metallicity exists for most neutron-captureelements. The large abundance scatter has been at-tributed to the elements being produced in more thanone type of nucleosynthesis process or astrophysicalenvironment. For example, an α element like Mg doesnot show a large star-to-star scatter, but only spreadsaround a mean value of ∼ . ± . 24 dex, while manyneutron-capture elements show a scatter of ± ∼ . Fig. 6.— Top: [Mg/Fe], middle: [Sr/Fe], and [Ba/Fe], bottom:[Eu/Fe]—all plotted versus [Fe/H]. A gray band around the averageabundance indicates the star-to-star abundance scatter (standarddeviation) for the full sample, while the dashed lines show thestandard deviation for our sample. The values are given in Table2. for a few selected elements. Just by applying the selec-tion criteria the star-to-star scatter has decreased (seeTable 2) and the selection criteria is therefore successfulin removing most of the contamination originating frommultiple processes as well as CEMP stars.To fully understand the remaining star-to-star scat-ter present in the reduced stellar sample, we rewrite thestandard [X/Fe] abundance convention to only includethe contribution from either the L component ([X L */Fe]– Eq. 4) or the H component ([X H */Fe] – Eq. 5). Thefirst logarithmic term in those equations corresponds to Fig. 7.— Individual process contribution from the H componentand L component to [Sr*, Zr*, Ba*, and Eu*/Fe] as a function of[Fe/H] from the top to the bottom. The L component is shown asfilled blue diamonds, while the H component is depicted as filledred squares. the total (normal) log ǫ ( X ).[ X ∗ L / Fe] = log(10 log ǫ ( X ) − ( Y H ( X ) · C H · [Fe / H] )) − log ǫ ( X ) ⊙ − [Fe / H] (4)[ X ∗ H / Fe] = log(10 log ǫ ( X ) − ( Y L ( X ) · C L · [Fe / H] )) − log ǫ ( X ) ⊙ − [Fe / H] (5)The L- and H-component contributions to the totalabundance are shown in Fig. 7 as a function of [Fe/H].Starting with the top panel showing [Sr ∗ /Fe], we con-clude that in most cases the L component contributesthe most to the Sr abundance derived from observationsof these stars. Only in a few stars, the H componentcontribution is larger. Table 3 shows the mean and stan-dard deviation from the mean for the elements shownin Fig. 7. While Zr behaves similar to Sr (L compo-nent creating most of the observed abundance), both Baand Eu are dominated by the H component. A largescatter is found for Ba and Eu vs. [Fe/H] and it in-dicates that these two n-capture H-component elementsare not co-produced with Fe (as originally postulated inSpite & Spite 1978). The standard deviation increases TABLE 2The mean and standard deviation for our sample and the full un-cut sample. < [Mg/Fe] > < [Sr/Fe] > < [Zr/Fe] > < [Ba/Fe] > < [Eu/Fe] > Sample 0.34 ± ± ± ± ± ± ± ± ± ± TABLE 3Mean ± standard deviation [Sr ∗ /Fe] [Zr ∗ /Fe] [Ba ∗ /Fe] [Eu ∗ /Fe]L − . ± . 32 0 . ± . − . ± . − . ± . − . ± . − . ± . − . ± . 60 0 . ± . the more the formation process of the element in the nu-merator differs from that of the element in the denomi-nator.The situation for the lighter elements, Sr and Zr, is lessclear cut because the larger scatter in these elements has the same size as the adopted uncertainty ( ± < Z < 50) or that the processes are less ro-bust.The robustness of the L- and H-component contribu-tions can be studied in Fig. 8 and Fig. 9. Barium and Eushow an almost perfect correlation for the H componentwith little spread ( ± . 19 dex). This confirms the robust-ness of the H component, which is in good agreementwith the universality of the r process (the most likely pro-cess behind the H component) shown in many other stud-ies (e.g. Cowan et al. 2002; Sneden et al. 2003). Stron-tium and Zr, on the other hand, show a larger star-to-star scatter in both L and H components, with standarddeviations of 0.27 dex and 0.38 dex, respectively. Eventhough the deviations are at the limit of the compo-nent uncertainties ( ± ≥ ± . 32 dex un-certainty, or that there may be additional nucleosynthe-sis components/processes hiding wihtin this uncertainty.The method and current level of abundance accuracy donot allow us to distinguish between these possibilities.By extracting stars from our sample with large L-component coefficients, we find strongly L-component-enriched stars to further test the robustness of the pro-cess. Figure 10 shows the stellar abundances of HE0104-5300, HE0340-5355, BS16469-075, HE1252-075, HE2219-0713, BD+4 2621, HD4306, HD88609, and HD122563which have a predominant L-component contribution.They show a larger spread in their abundance patternthan first expected (see Fig. 3, where the well-knownL-component stars agree within ± . even the component separated abundances Fig. 8.— The calculated contribution from the L component ( Y L blue, diamonds) and the main H component ( Y r red, squares) forSr and Zr. Fig. 9.— The L- and H-calculated abundances for Eu and Ba.Details can be found in the legend. ment). The larger sample of possible L-dominated starsindicate that these show an abundance spread within ± . 32 dex, which is the allowed abundance uncertaintyadopted for our pattern fitting (see Fig. 10). The in-crease in the abundance spread could indicate we needto classify the L-component stars such as HD122563 andHD88609 better, to distinguish between such stars andthe others shown in Fig. 10. Alternatively, the L-dominated stars actually span a wider range of abun-dances than H-dominated (e.g., the r-II) stars do, andthe assumed robustness of the L component may breakdown. If this is the case, this would allow for a widerrange of astrophysical conditions facilitating the processor for several processes to coexist and blend into our Lcomponent thereby increasing the scatter to 0.32 dex. Fig. 10.— Abundances normalized to Sr for L-component starsand candidate stars with promising L-component patterns. Recently, Roederer (2013) showed an almost perfectcorrelation between Sr and Ba and concluded that allstars must have heavy elements in their atmospheres.Thereason why we do not detect them is due to weak linesand observational biases. In Fig. 11, we show that bothSr and Ba grow ( as in, e.g., Aoki et al. (2005); Roederer(2013)), almost at the same rate. This could indicate acoproduction of Sr and Ba in the same site or that thenucleosynthesis processes create both elements in almostthe same amounts. However, the large star-to-star scat-ter could veil differences in the formation process and/orsite between Sr and Ba. Accurate isotopic abundancesfor both elements in a large sample are needed to settlethis issue. Currently, only Ba has been studied on anisotopic level in small samples (see, e.g., Roederer et al.2008; Gallagher et al. 2012 and references therein). Onlywhen splitting the Sr and Ba abundances into compo-nents we detect the differences in the single formationprocesses (for comparison see Figs. 8 and 9). This sep-aration method is currently the best proxy for isotopicabundances. We clearly see the difference in how Sr ispredominantly produced by an L-component process, aprocess which does not have to produce any or only lit-tle Ba (see Figs. 7–11). This separation into L- and H-components partially detaches the origin of these twoelements as they can be produced in different ratios ineach component/process. The relation showed in Fig. 11indicates that despite the fact that all stars have mostlikely been enriched in heavy elements, they do not needto be created by the same process, but could originatefrom the same object via different processes or have beenmixed from different sites prior to incorporation. Thisstatement will need to be verified in the future by im-proved yield predictions as well as GCE models, which will help us disentangle the formation sites and physicalquantities. Fig. 11.— Log ǫ abundances of Sr vs. Ba. The contributionsfrom the L component (diamonds) and the H component (main rprocess) are over plotted (squares). Predicting abundances Europium is an important element because it is almosta pure r-process tracer. Thus, we need to know how thiselement behaves observationally at the lowest metallici-ties to constrain and optimize theory. Figure 12 showsthat it is very challenging (or impossible) to measureEu abundances in extremely and hyper metal-poor stars(which confirms the bias mentioned in Roederer 2013).This is a problem because the Eu abundances are neededto compute abundance patterns and to improve resultsand interpretations from, e.g., galactic chemical evolu-tion (GCE) models. These models rely on the number(statistics) of the observationally derived abundances.Our method may help improve the statistics for futureGCE models of Eu.Since 94% of Eu is created by the r process/H com-ponent (see e.g., Bisterzo et al. 2014), one can use thedescribed method (Sect. 3) to predict Eu abundances,which cannot be derived from observations owing to thevery weak absorption lines found at low-metallicity (orpoor quality) spectra. By assuming that Eu is purelycreated in the H component, Eu can be calculated via ascaling relation between the H component’s Eu and Baabundances. Barium is an obvious choice because thiselement is a good H-component tracer at low metallic-ity and it shows fairly strong absorption lines even inextremely metal-poor stars. Therefore, we know the Baabundance in a much larger number of stars than theones for which we know the Eu abundance (see Fig. 6).The relation between the H component Eu and Ba (see0 Fig. 12.— Three synthetic spectra of giant stars with the samestellar model parameters (T=5200K, log g = 2.0), but three differ-ent metallicities: [Fe/H]= − , − 2, and − − . ≤ − Fig. 13.— H-component fraction of the log ǫ abundances of Baand Eu for the selected sample (filled red squared), for all stars withcalculated components (black plusses), and predicted Eu abun-dances (green triangles). Lines have been fitted to the selectedsample as well as all stars with calculated components. Fig. 13) can be then used to predict ‘unobserved’ Euabundances (triangles in the figure):log ǫ (E u ) r = 0 . · log ǫ (Ba) r − . . (6)By using this relation, the number of Eu abundancesknown below [Fe/H] = − . Fig. 14.— [Eu/Fe] as a function of [Fe/H]. (Observationally de-rived Eu shown as pluses.) The predicted, i.e., not observationallydetectable main H component [Eu/Fe] vs, [Fe/H] (green triangles).The yellow band is the standard deviation as shown in Fig. 6. able Eu abundances, which for most stars are trustwor-thy, but it cannot predict accurate abundances for starssuch as the outliers mentioned in Sect. 3.2.Although, this method can by no means replace realobservationally derived abundances, it can be used toestimate, e.g., Eu abundances either for GCE calcula-tions or to estimate stellar abundances when applyingfor follow-up observing time. Using components to constrain astrophysics In this section, we use the derived L-component abun-dances to constrain the astrophysical conditions of oneof the possible sites where this component may be pro-duced: neutrino-driven winds in core collapse super-novae. Similar studies have been carried out for the rprocess (see, e.g., Mumpower et al. (2012)).Neutrino-driven winds occur after a successful core-collapse supernova explosion, when neutrinos deposittheir energy in the outer layers of the neutron stars,and this layer gets ejected (see Arcones & Thielemann(2013) for a recent review). Although neutrino-drivenwinds were thought to be the site for the r process(Woosley et al. 1994), recent hydrodynamic simulationshave shown that the required extreme conditions are notreached. It is still possible that the winds may have theconditions necessary to produce the L-component condi-tions as have been explored in Arcones & Montes (2011).To explore under which astrophysical conditions inneutrino-driven winds are capable of reproducing the L-component abundances, we have systematically modifieda wind trajectory from Arcones et al. (2007). This tra-jectory corresponds to an explosion of a 15 M ⊙ progeni-tor based on Newtonian hydrodynamic simulations witha simple neutrino transport. The simplification in thetransport makes it possible to study the evolution from1 40 50 60 70 80 90 100 110Entropy [k B /nuc]0.400.420.440.460.480.500.520.540.560.580.600.620.64 Y e log( Y(Sr)/Y(Y) ) −1.0−0.50.00.51.01.52.040 50 60 70 80 90 100 110Entropy [k B /nuc]0.400.420.440.460.480.500.520.540.560.580.600.620.64 Y e log( Y(Sr)/Y(Zr) ) −2−101240 50 60 70 80 90 100 110Entropy [k B /nuc]0.400.420.440.460.480.500.520.540.560.580.600.620.64 Y e log( Y(Sr)/Y(Ag) ) Fig. 15.— Ratios of abundances for Sr and Y (upper panel), Srand Zr (middle panel), and Sr and Ag (bottom panel) are shownin logarithmic scale for different entropies and Y e . In the hatchedregions the ratios agree with the ones of the L component ± a few milliseconds up to few seconds post bounce for var-ious progenitors and explosion energies in one and twodimensions (for more details see Arcones et al. (2007);Arcones & Janka (2011)). The approximations in thesimulations may lead to small variations of the wind pa-rameters such as expansion timescale, entropy, and elec-tron fraction compared to what was obtained in the orig-inal trajectory. In order to account for the uncertaintyin the wind parameters, we systematically varied themwithin their expected uncertainty. To explore differententropies ( S ∝ T /ρ ), the density was reduced and in-creased within ∼ ± 40 50 60 70 80 90 100 110Entropy [k B /nuc]0.400.420.440.460.480.500.520.540.560.580.600.620.64 Y e Fig. 16.— L component predicts the ratio of the abundances forSr/Y, Sr/Zr, and Sr/Ag within some error bars. This figure showsthe wind parameter space and the regions where the ratios Sr/Y( // ), Sr/Zr ( \\ ), and Sr/Ag (green) agree with the L-componentpredictions. Figure 15 shows abundance ratios of Sr with respect toY, Zr, and Ag using a wind trajectory ejected five sec-onds after bounce as a function of entropy and electronfraction. The calculated L-component ratios shown inFig. 15 are: Sr/Y=6.13, Sr/Zr=1.22, and Sr/Ag=48.25.These values with their associated observational and sta-tistical errors of the method ( ± . 32 dex) correspond tothe marked regions in the figures. Figure 16 shows theoverlapping regions where all the simulated ratios (Sr/Y,Sr/Zr, and Sr/Ag) agree with the extracted L-componentratios. For very proton-rich conditions, there is a bandwhere all ratios overlap. Such conditions ( Y e ≈ . S ≈ k B / nuc) may be achieved a few secondsafter the explosion. Note that the abundances in proton-rich winds depend on the electron antineutrino lumi-nosity and energy because the nuclei are produced bythe νp process (Pruet et al. 2006; Fr¨ohlich et al. 2006;Wanajo 2013). Slightly different antineutrino luminosi-ties and energies may result in different entropies andelectron fractions. Figure 16 also shows slightly differentproton-rich conditions ( Y e ∼ . − . 55) that could re-produce the derived L-component abundances. However,this slightly proton-rich ( Y e ∼ . − . 55) region of theparameter space is less relevant because the abundancesof elements between Sr and Ag for such conditions arevery small. In neutron-rich ( Y e < . 5) conditions theL-component ratios can be reproduced only in a verynarrow band of the parameter space. For small vari-ations of the wind parameters the abundances changesteeply in contrast to the smooth trend in proton-richwinds (Fig. 15 and Arcones & Bliss (2014)). During thetime evolution of the wind, the parameters are likely toevolve and change. Therefore, given the very narrowrange of neutron-rich conditions that can reproduce theL abundances, it is likely that a small change in the as-trophysical conditions will result in a non-robust L com-ponent. If, on the other hand, the L component is ro-2bust (as assumed in this paper), proton-rich conditionsare more feasible, since they allow for a wider range ofastrophysical conditions and still explain the observed Labundances. All previous discussion is based on a singleexpansion timescale, we have also studied other trajec-tories varying this quantity and find that the qualitativebehavior and conclusions are the same.The parameters of the neutrino-driven wind evolvewith time as the neutrino luminosity decreases and theneutron star contracts and cools. Variations are alsoexpected for different stellar progenitors because moremassive ones will lead to more massive neutron starswhich, in turn, means higher entropy allowing heavierelements to form (Qian & Woosley 1996; Otsuki et al.2000; Thompson et al. 2001). Therefore, the contribu-tion from neutrino-driven winds to the observed abun-dances and to the L component comes from combina-tions of wind parameters, i.e., various points in Figs. 15–16. In order to explain the L component, most of themass needs to be ejected with parameters from the over-lapping regions in Fig. 16, even if any combination ofwind parameters can be realized. If the contribution fromneutrino-driven winds comes from very different regionsof the parameter space, the abundance pattern would notreproduce the L component and it would not be robust.Therefore, the robustness (within error bars) stronglyconstrains the astrophysical conditions where the L com-ponent is produced.Which is the heaviest element that can be produced inneutrino-driven winds for typical wind parameters? Fortypical wind conditions, Ba is not produced or only ina negligible amount. Therefore, if observations confirmthat the L component extends up to Ba, the neutrino-driven wind (as obtained in current hydrodynamic sim-ulations) is not likely to be responsible for the L compo-nent. SUMMARY & CONCLUSION Abundances from metal-poor stars provide clues aboutthe origin of the elements in the early universe. We haveused the large inhomogeneous sample of metal-poor starsfrom Frebel et al. (2010), after applying selection crite-ria to remove contamination from s-process abundances,self mixing, etc., to obtain abundances from single nu-cleosynthesis processes. By assuming only two nucle-osynthesis processes (the H and L components followingQian & Wasserburg 2001) contribute to the metal-poorstellar abundances, and that each single event createsrobust abundances (within ± . ± . ± . Z ≥ 56 are gen-erally created by the H component, while the abundanceof elements 38 ≤ Z ≤ 47 created by a combination ofL and H components. Since the L component does nothave a significant contribution to elements Z ≥ 56, theexact L abundance to those elements (obtained by thedifferent methods) does not change or affect the goodagreement between our model and observationally de-rived abundances. In each method, we found one or afew outlying stars that could not be explained under ourassumptions using this method. This could indicate thatour method is incomplete or that our assumptions aretoo strong. However, these outliers (in model 4) seemto have been self-enriched due to internal mixing, whichcauses the poor χ values. The general good agreementbetween our simple analytical model and the metal-poorstellar abundances indicates that two robust nucleosyn-thesis processes are responsible for the abundances ofalmost all metal-poor stars considered here.By deconvolving the stellar abundances into the indi-vidual components, we have also studied the star-to-starabundance scatter. Since the H component is mainlyresponsible for the Z ≥ 56 abundances, most of the ob-served scatter, as a function of metallicity is also observedin the H contribution, and it indicates that Fe and theH component are not co-produced (as originally postu-lated in Spite & Spite 1978). The robustness of the H-component contributions was confirmed by studying thescatter between different H contributions ( < ± . ≤ Z ≤ 47, the situationis not as clear since the observed scatter as a functionof metallicity in the L component is similar to the totalattributed uncertainty in our model ( ± . 32 dex). Thescatter between different L-elemental contributions wasalso at the limit of what was expected ( ± . 32 dex). This,and the fact that the abundance pattern of stars witha relative large L-component contribution is larger thanoriginally expected, suggests that the H component has asmaller intrinsic scatter compared to the L component’sintrinsic scatter. The relative larger scatter of the L com-ponent may either indicate that the process is robust onlywithin a ± . 32 dex uncertainty or that there could be ad-ditional nucleosynthesis components or processes hidingwithin the allowed uncertainty. The method used in thispaper does not allow us to distinguish between these pos-sibilities.We consider neutrino-driven winds in core-collapsesupernovae as a possible site and use the derived L-component abundances to constrain the astrophysicalwind conditions. In order to explain these abundances,the environment likely needs to be proton-rich within asignificantly constrained parameter space. If the L com-ponent is not robust, or if a single event has evolving con-ditions, there are even narrower bands of the parameterspace that can reproduce the observations. In addition,the neutrino-driven winds are not likely to create Ba orheavier elements. If future abundance observations of L-dominated stars show that Ba is likely to be producedin an L component, their abundances are not likely tobe created in neutrino-driven winds in core-collapse su-pernovae. Both simulations and observations need to3improve to further constrain the production of heavy ele-ments and assess the robustness and number of processesworking at low metallicity.C. J. Hansen was supported by Sonderforschungsbere-ich SFB 881 ”The Milky Way System” (subproject A5)of the German Research Foundation (DFG) and the VIL- LUM Foundation. F. Montes was supported by the JointInstitute for Nuclear Astrophysics at MSU under NSFPHY grant 08-22648. The work of A. Arcones was sup-ported by the Helmholtz-University Young Investigatorgrant No. VH-NG-825. We also wish to thank the INTfor providing a fruitful forum for discussions (INT-PUB-14-044). APPENDIX UNCERTAINTIES FROM OBSERVATIONS AND MODELS To assess how accurate the method is in splitting the abundance contributions from the L and H components, wecarried out several tests to answer two questions: (1) Does the outcome depend on the intial set of stars? (2) Howdo the observational uncertainties (either intrinsic or due to the inhomogeneous stellar sample) affect the derivedabundances? For this purpose, we use a new set of stars that are neither predominantly L- or H-component enriched,and therefore we used method M3 because it does not assume a dominant process. BD+4 2621 and HD6268 wereselected because these stars have detailed abundances for a large number of elements. The derivation of the L- andH-component abundances using these stars’ abundances is referred to as method M3b. The impact of the uncertaintieson the observed stellar abundances and the inhomogeneity of the sample was tested by creating an extreme case basedon BD+4 2621 (method M3c). The new abundances in this “modified” star were “created” by randomly selectingabundances for that specific star found in literature for each element (generally the literature values vary within theobservational uncertainty). The abundances of other stars were not modified. Figure 17 shows the abundances obtainedby using BD+4 2621 and HD6268 in method M3b, and using method M3c for the “modified” BD+4 2621 and HD6268.Similar results are obtained using different combinations of stars. All H- and L-component abundances in the range38 ≤ Z ≤ 47 are very similar and show a good agreement within ± ± Z ≥ 56, the differences among the methods(M3, M3b, and M3c) are larger and this reflects the intrinsic limitations of the method (or physical features of theprocess, since the L component may not create these nuclei very efficiently). The more similar the abundances of thestars are, the larger the uncertainty in the decomposed abundances get. The difference between the methods can betaken as a conservative uncertainty in the component abundances, because BD+4 2621 are HD6268 are not dominatedby an individual component (L or H). The uncertainty obtained based on M3, M3b , and M3c is similar to the onebased on M1, M2, and M3. Therefore, the use of different L-component abundances (M1 and M2 in Table 1) alsocovers the error due to the choice of the initial stars. Z35 40 45 50 55 60 65 70 75 l og -3-2-101 M1M2M3M3bM3c Z35 40 45 50 55 60 65 70 75 l og -3-2-10 M1,M2M3M3bM3c ∈∈ Fig. 17.— M1, M2, M3 refers to the fitting methods described in Sect. 3 and the letters b and c indicate that BD+4 2621 and HD6268were used instead of HD122563 and CS 22892-052. Since the uncertainty for both components (excluding L component Z ≥ 56) is within ∼ . ± . 32 dex, and it is used in the abundance deconvolution described in the Sect. 3.2. REFERENCESAnders, E. & Grevesse, N. 1989, Geochim. Cosmochim. Acta, 53,197Andrievsky, S. M., Spite, F., Korotin, S. A., Fran¸cois, P., Spite,M., Bonifacio, P., Cayrel, R., & Hill, V. 2011, A&A, 530, A105Andrievsky, S. M., Spite, M., Korotin, S. A., Spite, F., Fran¸cois,P., Bonifacio, P., Cayrel, R., & Hill, V. 2009, A&A, 494, 1083Aoki, W., Beers, T. C., Christlieb, N., Norris, J. E., Ryan, S. G.,& Tsangarides, S. 2007, ApJ, 655, 492Aoki, W., Honda, S., Beers, T. C., Kajino, T., Ando, H., Norris,J. E., Ryan, S. G., Izumiura, H., Sadakane, K., &Takada-Hidai, M. 2005, ApJ, 632, 611Arcones, A. & Bliss, J. 2014, Journal of Physics G NuclearPhysics, 41, 044005Arcones, A. & Janka, H.-T. 2011, aap, 526, A160+Arcones, A., Janka, H.-T., & Scheck, L. 2007, aap, 467, 1227Arcones, A. & Montes, F. 2011, ApJ, 731, 5Arcones, A. & Thielemann, F.-K. 2013, Journal of Physics GNuclear Physics, 40, 013201Arnould, M., Goriely, S., & Takahashi, K. 2007, Phys. Repts.,450, 97Barklem, P. S., Christlieb, N., Beers, T. C., Hill, V., Bessell,M. S., Holmberg, J., Marsteller, B., Rossi, S., Zickgraf, F.-J., &Reimers, D. 2005, A&A, 439, 129Bauswein, A., Goriely, S., & Janka, H.-T. 2013, ApJ, 773, 78Beers, T. C. & Christlieb, N. 2005, ARA&A, 43, 531Bergemann, M., Hansen, C. J., Bautista, M., & Ruchti, G. 2012,A&A, 546, A90Bisterzo, S., Gallino, R., Straniero, O., Cristallo, S., & K¨appeler,F. 2010, MNRAS, 404, 1529—. 2011, MNRAS, 418, 284Bisterzo, S., Travaglio, C., Gallino, R., Wiescher, M., & K¨appeler,F. 2014, ApJ, 787, 10Burbidge, E. M., Burbidge, G. R., Fowler, W. A., & Hoyle, F.1957, Rev. Mod. Phys., 29, 547Burris, D. L., Pilachowski, C. A., Armandroff, T. E., Sneden, C.,Cowan, J. J., & Roe, H. 2000, ApJ, 544, 302Cowan, J. J., Burris, D. L., Sneden, C., McWilliam, A., &Preston, G. W. 1995, ApJ, 439, L51Cowan, J. J., Sneden, C., Burles, S., Ivans, I. I., Beers, T. C.,Truran, J. W., Lawler, J. E., Primas, F., Fuller, G. M., Pfeiffer,B., & Kratz, K.-L. 2002, ApJ, 572, 861Cruz, M. A., Serenelli, A., & Weiss, A. 2013, A&A, 559, A4Farouqi, K., Kratz, K.-L., Pfeiffer, B., Rauscher, T., Thielemann,F.-K., & Truran, J. W. 2010, ApJ, 712, 1359Fran¸cois, P., Depagne, E., Hill, V., Spite, M., Spite, F., Plez, B.,Beers, T. C., Andersen, J., James, G., Barbuy, B., Cayrel, R.,Bonifacio, P., Molaro, P., Nordstr¨om, B., & Primas, F. 2007,A&A, 476, 935Frebel, A., Simon, J. D., Geha, M., & Willman, B. 2010, ApJ,708, 560Freiburghaus, C., Rosswog, S., & Thielemann, F.-K. 1999, ApJ,525, L121Frischknecht, U., Hirschi, R., & Thielemann, F.-K. 2012, A&A,538, L2Fr¨ohlich, C., Mart´ınez-Pinedo, G., Liebend¨orfer, M., Thielemann,F.-K., Bravo, E., Hix, W. R., Langanke, K., & Zinner, N. T.2006, Physical Review Letters, 96, 142502Fulbright, J. P. 2002, AJ, 123, 404Gallagher, A. J., Ryan, S. G., Hosford, A., Garc´ıa P´erez, A. E.,Aoki, W., & Honda, S. 2012, A&A, 538, A118Hansen, C. J., Andersen, A. C., & Christlieb, N. 2014, A&A, 568,A47Hansen, C. J., Bergemann, M., Cescutti, G., Fran¸cois, P.,Arcones, A., Karakas, A. I., Lind, K., & Chiappini, C. 2013,A&A, 551, A57Hansen, C. J., Primas, F., Hartman, H., Kratz, K.-L., Wanajo, S.,Leibundgut, B., Farouqi, K., Hallmann, O., Christlieb, N., &Nilsson, H. 2012, A&A, 545, A31Hill, V., Plez, B., Cayrel, R., Beers, T. C., Nordstr¨om, B.,Andersen, J., Spite, M., Spite, F., Barbuy, B., Bonifacio, P.,Depagne, E., Fran¸cois, P., & Primas, F. 2002, A&A, 387, 560 Hoffman, R. D., Woosley, S. E., Fuller, G. M., & Meyer, B. S.1996, apj, 460, 478Honda, S., Aoki, W., Ishimaru, Y., & Wanajo, S. 2007, ApJ, 666,1189Honda, S., Aoki, W., Kajino, T., Ando, H., Beers, T. C.,Izumiura, H., Sadakane, K., & Takada-Hidai, M. 2004, ApJ,607, 474Hotokezaka, K., Kyutoku, K., Tanaka, M., Kiuchi, K., Sekiguchi,Y., Shibata, M., & Wanajo, S. 2013, ApJ, 778, L16Just, O., Bauswein, A., Ardevol Pulpillo, R., Goriely, S., &Janka, H.-T. 2014, ArXiv e-printsKorobkin, O., Rosswog, S., Arcones, A., & Winteler, C. 2012,MNRAS, 426, 1940Lai, D. K., Bolte, M., Johnson, J. A., Lucatello, S., Heger, A., &Woosley, S. E. 2008, ApJ, 681, 1524Lattimer, J. M. & Schramm, D. N. 1974, ApJ, 192, L145Li, H., Shen, X., Liang, S., Cui, W., & Zhang, B. 2013, PASP,125, 143Lucatello, S., Tsangarides, S., Beers, T. C., Carretta, E., Gratton,R. G., & Ryan, S. G. 2005, ApJ, 625, 825Mart´ınez-Pinedo, G., Fischer, T., Lohs, A., & Huther, L. 2012,Physical Review Letters, 109, 251104Masseron, T., Johnson, J. A., Plez, B., van Eck, S., Primas, F.,Goriely, S., & Jorissen, A. 2010, A&A, 509, A93McWilliam, A., Preston, G. W., Sneden, C., & Searle, L. 1995a,AJ, 109, 2757McWilliam, A., Preston, G. W., Sneden, C., & Shectman, S.1995b, AJ, 109, 2736Metzger, B. D. & Fern´andez, R. 2014, MNRAS, 441, 3444Montes, F., Beers, T. C., Cowan, J., Elliot, T., Farouqi, K.,Gallino, R., Heil, M., Kratz, K., Pfeiffer, B., Pignatari, M., &Schatz, H. 2007, ApJ, 671, 1685Mumpower, M. R., McLaughlin, G. C., & Surman, R. 2012, Phys.Rev. C, 85, 045801Otsuki, K., Tagoshi, H., Kajino, T., & Wanajo, S. 2000, ApJ,533, 424Perego, A., Rosswog, S., Cabez´on, R. M., Korobkin, O., K¨appeli,R., Arcones, A., & Liebend¨orfer, M. 2014, MNRAS, 443, 3134Pignatari, M., Gallino, R., Meynet, G., Hirschi, R., Herwig, F., &Wiescher, M. 2008, ApJ, 687, L95Pruet, J., Hoffman, R. D., Woosley, S. E., Janka, H.-T., & Buras,R. 2006, ApJ, 644, 1028Qian, Y. & Wasserburg, G. J. 2007, physrep, 442, 237—. 2008, ApJ, 687, 272Qian, Y.-Z. & Wasserburg, G. J. 2001, ApJ, 559, 925Qian, Y.-Z. & Woosley, S. E. 1996, ApJ, 471, 331Roberts, L. F., Reddy, S., & Shen, G. 2012, Phys. Rev. C, 86,065803Roederer, I. U. 2013, AJ, 145, 26Roederer, I. U., Cowan, J. J., Karakas, A. I., Kratz, K.-L.,Lugaro, M., Simmerer, J., Farouqi, K., & Sneden, C. 2010,ApJ, 724, 975Roederer, I. U., Lawler, J. E., Sneden, C., Cowan, J. J., Sobeck,J. S., & Pilachowski, C. A. 2008, ApJ, 675, 723Roederer, I. U., Preston, G. W., Thompson, I. B., Shectman,S. A., Sneden, C., Burley, G. S., & Kelson, D. D. 2014, AJ,147, 136Ryan, S. G., Norris, J. E., & Beers, T. C. 1996, ApJ, 471, 254Sneden, C., Cowan, J. J., & Gallino, R. 2008, araa, 46, 241Sneden, C., Cowan, J. J., Lawler, J. E., Ivans, I. I., Burles, S.,Beers, T. C., Primas, F., Hill, V., Truran, J. W., Fuller, G. M.,Pfeiffer, B., & Kratz, K.-L. 2003, ApJ, 591, 936Spite, M., Cayrel, R., Plez, B., Hill, V., Spite, F., Depagne, E.,Fran¸cois, P., Bonifacio, P., Barbuy, B., Beers, T., Andersen, J.,Molaro, P., Nordstr¨om, B., & Primas, F. 2005, A&A, 430, 655Spite, M. & Spite, F. 1978, A&A, 67, 23Stancliffe, R. J., Dearborn, D. S. P., Lattanzio, J. C., Heap, S. A.,& Campbell, S. W. 2011, ApJ, 742, 121Straniero, O., Cristallo, S., Gallino, R., & Dominguez, I. 2004,Mem. Soc. Astron. Italiana, 75, 6655