Fluorine in the Solar Neighborhood: The Need for Several Cosmic Sources

The final fluorine abundances and upper limits are given in Tables 2 and 3, where we provide the number density abundances, A(F) = log N_F/N_H + 12, and the [F/Fe]¹² abundance ratios. In the tables, we also provide the oxygen and cerium abundance results from the optical spectra.

In Figure 2 we present [F/H] versus [Fe/H] as determined from the IGRINS data (in red), the Phoenix data (in blue), and from the Jönsson et al. (2014b) TEXES determinations (in green). Upper limits are marked with triangles. In Figure 3, the [F/Fe] trend is shown instead. The stellar abundances are normalized to the solar value, which is very uncertain; there is no detectable HF in the solar photosphere, so the solar abundance value is determined either from meteoritic measurements (${A}_{\odot }({\rm{F}})=4.43\pm 0.06$; Lodders 2003) or from uncertain measurements in sunspot spectra (${A}_{\odot }({\rm{F}})=4.56\pm 0.30$; Hall & Noyes 1969; Grevesse et al. 2007).¹³ More recently, Maiorca et al. (2014) analyzed a spectrum of a medium-strong sunspot umbra (∼4250 K) determining the solar fluorine abundance of ${A}_{\odot }({\rm{F}})=4.40\pm 0.25$, which is consistent with the Hall & Noyes (1969) value, with an equally large uncertainty. This nominal value is very close to the meteoritic value, but given the uncertainties in the modeling of the umbral spectra, the value to be used is the meteoritic value of Lodders (2003). It would be desirable to determine the solar fluorine abundance to a higher accuracy.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As can be seen in Table 2, only the giants with an effective temperature of less than approximately 4500 K show the HF line. For the metal-poor stars, even cooler giants are required. The method that we use for determining the stellar parameters is developed for stars warmer than approximately 4000 K, which means that we have avoided cooler stars as we aim for high accuracy and homogeneity. We realize that the HF line should get stronger for cooler stars and for stars with lower surface gravities, and these could be used for future fluorine measurements in the metal-poor region (see, e.g., the recent work by Guerço et al. 2019a). Cool, metal-poor giants are, however, very rare.

Several effects could contribute to the uncertainties in the determined fluorine abundances. The uncertainties in the HF line data are very small compared to other uncertainties, at least by a factor of 10. Residuals from the telluric line division could impinge on the HF line. We have, however, checked where the telluric lines fall in all of the stars, ensuring that this is not the case. Furthermore, the way the continuum is set could affect the equivalent width of the line. We have, however, adjusted the continuum locally, minimizing this uncertainty. The largest uncertainties stem instead from the uncertainties in the stellar parameters of the stars, in spite of our efforts to determine these as accurately as possible. We have therefore allowed the stellar parameters to vary within their uncertainties. The molecular lines are very temperature sensitive, with a change in the derived fluorine of 0.1 dex for +50 K. A change of the surface gravity of 0.15 dex results in a change of the same magnitude. The uncertainties in abundance ratios are, in general, smaller as they often cancel out to various degrees. A change in the microturbulence does not affect the synthesized line, as expected for these weak lines. This shows that it is important to determine the T_eff and $\mathrm{log}g$ very well in order to minimize the scatter in the fluorine trends with metallicity. We estimate a total uncertainty in the derived fluorine abundances to be σA(F) ∼ 0.15 dex and in the abundance ratios to be σ([F/Fe]) ∼ 0.10 dex.

The robustness of our method is demonstrated by the similarity between our abundance trends derived from the two different spectrographs (IGRINS and Phoenix) and that determined from the TEXES spectrograph, showing the similarity of the abundances derived from the two diagnostics (the vibration–rotational lines at 2.3 μm and the pure rotational lines at 12 μm).

In the following, a short overview of the possible production sites of fluorine that may be important and are discussed theoretically is given. The predicted trends from the different channels are discussed, which will be important when analyzing our data later on.

(i) The contributions to the cosmic budget of fluorine from nonrotating massive stars and conventional Type II supernovae are negligible (Kobayashi et al. 2011b; Prantzos et al. 2018). Rapidly rotating massive stars can, however, produce primary fluorine from ¹⁴N, via proton and α captures in the presence of ¹³C, which is needed for the generation of protons (Prantzos et al. 2018). The ¹⁴N comes from reactions with ¹²C which is the ashes of He burning in the massive star itself and is therefore of primary origin (Guerço et al. 2019a). This process could dominate the F budget all the way up to solar metallicities.

(ii) The ν process, active during core-collapse supernovae (Woosley & Haxton 1988; Woosley et al. 1990; Timmes et al. 1995; Langanke et al. 2019), could contribute substantially to the fluorine production (Kobayashi et al. 2011a; Nault & Pilachowski 2013; Pilachowski & Pace 2015). In this process, fluorine is formed from neutrino-induced spallation reactions with ²⁰Ne in the expelled shell, containing nuclei formed in the progenitor star. Fluorine is then a primary element, formed from a process that is independent of the metallicity of the site of formation. Because the progenitor stars are massive, short-lived stars, they contribute early (Olive & Vangioni 2019), already at low metallicities, and at a constant ratio with oxygen, another element synthesized in massive stars ending their lives as Type II supernovae. However, there are still large uncertainties in the stellar modeling (such as progenitor mass and distribution of ²⁰Ne) and the neutrino-induced thermonuclear reaction rates of ²⁰Ne (Sieverding et al. 2018, 2019). The F production is also very sensitive to the modeled neutrino flux and its spectrum (Alibés et al. 2001; Prantzos et al. 2018; Langanke et al. 2019; Olive & Vangioni 2019).

(iii) During the He-burning thermal pulses (TP) in AGB stars, the ¹⁴N produced in the hydrogen-burning CNO cycle can produce fluorine through a chain of reactions also involving neutrons and protons, a process discussed by, for example, Forestini et al. (1992), Jorissen et al. (1992), Abia et al. (2011), and Cristallo et al. (2014). Subsequently, the star undergoes the third dredge-up, during which the surface is enriched with fluorine. This fluorine is subsequently expelled to the interstellar medium by stellar winds and/or during the planetary nebula phase of the star. Fluorine produced through this channel would be a secondary element (Prantzos et al. 2018), with yields depending on the metallicity of the star forming it. Goriely & Mowlavi (2000) also show the importance of partial mixing of protons into the carbon-rich layers in the interiors of AGB stars during the third dredge-up, for the formation of fluorine.

At too high temperatures in the stellar interiors, helium-nuclei or proton-capture reactions destroy fluorine, converting it to Ne, as mentioned above. Therefore, the AGB stars that form fluorine are less massive than approximately 4 ${M}_{\odot }$, preventing the high temperatures of hot bottom burning (Kobayashi et al. 2011a). The fluorine production in AGB stars is, indeed, mass dependent between 2 and 4 ${M}_{\odot }$, with a maximum around 3 ${M}_{\odot }$ (Kobayashi et al. 2011b). Due to the time delay for low- and intermediate-mass stars to start contributing to the cosmic buildup of fluorine, this source will start contributing at a higher metallicity ([Fe/H] ∼ −0.9 to −0.7 Kobayashi et al. 2011a) and will be different for different stellar populations with different star formation rates. This process was observationally argued to be a dominant source of present-day cosmic fluorine in the solar neighborhood by, for example, Jorissen et al. (1992), Recio-Blanco et al. (2012), Jönsson et al. (2014b), Olive & Vangioni (2019), Guerço et al. (2019a), and Abia et al. (2015a, 2015b). The last also concludes that additional sources are necessary. Abia et al. (2019) argue that AGB stars do contribute, not as the main source in the solar neighborhood, however, and thus contrary to Olive & Vangioni (2019), among others.

(iv) A similar reaction chain could occur in massive Wolf–Rayet stars, a process investigated early on by Meynet & Arnould (2000) and discussed and argued for in, e.g., Cunha et al. (2003), Renda et al. (2004), Cunha et al. (2008), and Spitoni et al. (2018), but questioned in Palacios et al. (2005) when rotation is included. ¹⁴N, also here resulting from the CNO cycle as a secondary element (N formed from the preexisting C), is converted to fluorine during the core-helium-burning phase, and subsequently expelled through the strong, metal-line driven wind. In this channel, the fluorine produced therefore acts as a secondary element, at high enough metallicities. The formation of Wolf–Rayet stars and their winds is also metallicity dependent, thus becoming increasingly important first for higher metallicities (Cunha et al. 2003) and should, therefore, be less dependent on the star formation rates of different stellar populations. Primary nitrogen, subsequently burning to fluorine, can also be produced, but is restricted to metal-poor, rotating massive stars (Meynet & Maeder 2002; Spitoni et al. 2018).

It should, however, be noted that the conditions and mechanisms in Wolf–Rayet stars for fluorine to survive and contribute to the cosmic F reservoir are very uncertain; recent massive-star models with mass-loss rates accounting for the reduction factor due to clumping show that these stars might not at all be significant as a source of cosmic fluorine (G. Meynet 2020, private communication). This was also shown earlier by Palacios et al. (2005).

(v) Spitoni et al. (2018) discuss the possibility of novae contributing to the cosmic budget of fluorine. These could, in principle, produce fluorine (José & Hernanz 1998) through reactions starting with proton captures by ¹⁷O nuclei. Indeed, Spitoni et al. (2018) conclude that this process might be important to reproduce the secondary behavior of the observed fluorine trends. It should also be noted that the novae yields are also highly uncertain (Spitoni et al. 2018).

The abundances as a function of metallicity, presented in Figure 2, show a very tight relation, close to linear. At lower metallicities, these abundance ratios might decrease more slowly as the metallicity decreases. The spread in the F abundances in the range between $[\mathrm{Fe}/{\rm{H}}]\sim -0.6$ up to solar metallicity is larger than that for supersolar metallicities. This is even more evident in Figure 3, where [F/Fe] is plotted versus metallicity. Below $[\mathrm{Fe}/{\rm{H}}]\sim -0.6$, we have three upper limits and one determined abundance from a clearly detected HF line in the spectrum of the giant HIP 63432, with $[\mathrm{Fe}/{\rm{H}}]=-0.87$. This spectrum is displayed in Figure 1 and has a [F/Fe] = 0.2. Note that our estimated uncertainties, mainly due to the uncertainties in the stellar parameters, is of the order of $\sigma [{\rm{F}}/\mathrm{Fe}]\sim 0.1$ dex. This low-metallicity trend with metallicity seems, thus, to be quite flat, with indications of a detectable spread. However, more fluorine determinations in metal-poor stars are clearly needed.

All trends go roughly through the expected solar values, ([F/Fe], $[\mathrm{Fe}/{\rm{H}}]$) = (0, 0). It should, however, be kept in mind that the solar F-abundance determinations are subjected to large uncertainties (of the order of ±0.25 dex; Hall & Noyes 1969; Grevesse et al. 2007; Maiorca et al. 2014) and that we are using the meteoric value of Lodders (2003). The choice of the value of the solar fluorine abundance only results in a scaling of the abundances in the figures, but should be taken into account for comparisons with other published abundances.

Our abundances show less scatter in the trends than many previous investigations (see, e.g, Recio-Blanco et al. 2012; Pilachowski & Pace 2015; Jönsson et al. 2017b). For the narrow metallicity range of the open cluster giants in Nault & Pilachowski (2013), our results agree very well, with a clear increase at supersolar values; see Figure 4. Furthermore, our new abundances based on the Phoenix spectra show a slightly smaller scatter, and a tighter correlation, than those determined by Jönsson et al. (2017b), who analyzed the same spectra (see the small gray dots in Figure 4). We believe the reason for this smaller scatter is caused by our more accurate stellar parameters. Also, the trends discussed in Guerço et al. (2019a) agree very well with ours (crosses in Figure 4), within the common metallicity range. In light of their plateau suggested by the data of Guerço et al. (2019a) at low metallicities ($[{\rm{Fe}}/{\rm{H}}]\lt -0.5$; excluding the probable members of the Monoceros overdensity) and the data point from the study of Li et al. (2013), our indications of a plateau is strengthened. The plateau in Guerço et al. (2019a), however, lies at a ∼0.2 dex lower level.

Zoom In Zoom Out Reset image size

In Figure 5, we plot the [F/Fe] abundance ratios as a function of effective temperature, color-coding the stars for their metallicities. Stars cooler than approximately 4350 K show no trend with T_eff but a scatter. For the warmer stars, only higher [F/Fe] values are measured, tracing the upper range of the scatter. This effect is also seen and discussed in Pilachowski & Pace (2015). In the warmer stars, the HF line increasingly disappears, and only the stars with a high F abundance will be detected. Lower abundance, such as for metal-poor stars, will not yield a strong-enough HF line to be detected for the warm stars. Indeed, among the warmer stars, we find mostly metal-rich stars, for which we see an upturn in [F/Fe] in Figure 3. Furthermore, few metal-poor stars are found at temperatures above 4350 K.

Zoom In Zoom Out Reset image size

In Figure 6, we plot the [F/Fe] abundance ratios instead, as a function of metallicity, color-coding the stars for their effective temperatures. No trend with T_eff is detected, apart from the most metal-rich stars being predominately warm. The reason for this is seen in the Kiel diagram in the lower-left corner of the figure. There, we clearly see the giant branch with the metallicities increasing diagonally to lower temperatures, as expected (see also Jönsson et al. 2017a). Most of our warm stars above 4400 K are red clump stars and are metal rich. Note that some of the high-metallicity stars are also cooler. Had we targeted more stars, we would have detected the HF line in a larger number of cooler stars. The line is actually more easily detected the cooler the giant star is, due to the molecular equilibrium. However, the lower number of warm, metal-poor stars in the Kiel diagram is caused by the HF line strength becoming weaker and eventually disappearing.

Zoom In Zoom Out Reset image size

In Figure 7, we again plot the [F/Fe] ratios as a function of metallicity, but this time color-coding the stars according to the assigned stellar population, i.e., thin-disk, thick-disk, or halo stars as determined from the separation in the [Mg/Fe] versus [Fe/H] plane (see H. Jönsson et al. 2020, in preparation for details). If anything, there is a slight tendency that the thick-disk stars seem to lie at the upper envelope of the thin-disk stars. More data are needed to confirm this observation. This is, however, theoretically expected according to the models in Kobayashi et al. (2011a, 2011b), also plotted in Jönsson et al. (2017b). We have also been able to determine an upper limit of the F abundance for the only halo star in our sample, which is, on the other hand, much lower than what is predicted in Kobayashi et al. (2011b).

Zoom In Zoom Out Reset image size

Another way to investigate the importance of the different nucleosynthetic processes for the cosmic fluorine budget is to look for the signatures of fluorine's primary or secondary origin; see Section 5.1. Because oxygen is produced as a primary element in massive stars (see, e.g., Prantzos et al. 2018), plotting the number density of fluorine as a function of the number density of oxygen will reveal whether fluorine is primary or whether it behaves like a secondary element. In Figure 8, we plot A(F) versus A(O). A linear regression gives a slope of 1.8 ± 0.2, which is a clear indication of a secondary behavior. One can also directly plot the fluorine-to-oxygen abundance ratio as a function of the oxygen abundance, which will be a constant for a primary element and give a slope of one for a secondary element. In Figure 9, this is plotted and we can reject the hypothesis of a primary element as the trend is not constant. The slope is instead close to one, 1 ± 0.2, which is the signature of a secondary element.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The synthesis of fluorine in evolved red giant stars has been demonstrated observationally by a number of studies, beginning with that of Jorissen et al. (1992) and more recently those of, e.g., Pandey (2006), Alves-Brito et al. (2011), Lucatello et al. (2011), and Abia et al. (2015a, 2019). This is the only nucleosynthetic process forming fluorine that has an observational proof of its activity. From several papers on the galactic chemical evolution of fluorine, the AGB-star contribution in the approximate metallicity range of $-0.6\lt [\mathrm{Fe}/{\rm{H}}]\lt 0.0$ is, indeed, claimed to be dominant (e.g., Renda et al. 2004; Recio-Blanco et al. 2012; Spitoni et al. 2018; Olive & Vangioni 2019). It should, however, be noted that there are different theoretical predictions of the AGB-star contributions, mainly due to which nuclear reaction rates are adopted. For instance, Prantzos et al. (2018) find a lower AGB stellar yield than Kobayashi et al. (2011b) or Spitoni et al. (2018); see also the discussion in Guerço et al. (2019a).

Due to the time delay of the AGB stars producing cosmic fluorine, their contribution is theoretically predicted to start contributing at metallicities larger than $[\mathrm{Fe}/{\rm{H}}]\sim -0.9$ for the thin disk and at $[\mathrm{Fe}/{\rm{H}}]\sim -0.7$ for the thick disk (see Kobayashi et al. 2011a, 2011b; Spitoni et al. 2018). Thus, with these theoretical predictions and our observationally determined secondary behavior, we conclude that the AGB-star contribution must be dominant at least in the $-0.6\lt [\mathrm{Fe}/{\rm{H}}]\lt 0.0$ range, thus corroborating the findings of Recio-Blanco et al. (2012), Jönsson et al. (2017b), Olive & Vangioni (2019), and Guerço et al. (2019a). The AGB contribution would also naturally cause the large spread in the [F/Fe] versus [Fe/H] plot, which we see in our data, due to the range in masses (and therefore time delays) contributing.

For high metallicities, above solar, we find an increasing [F/Fe] trend displaying only a small scatter. This part also shows a secondary behavior. The large increasing trend is surprising, because according to galactic chemical evolution models, like the ones in Spitoni et al. (2018), the Fe contribution from Type Ia supernovae should nominally decrease the trend. The yields needed to increase it instead are larger than the current best description of the yields from Wolf–Rayet stars and novae (Spitoni et al. 2018), which are, most likely, even overestimated. In order to fit the increasing abundances at high metallicities, Spitoni et al. (2018) tried to model the galactic chemical evolution of fluorine by artificially increasing the Wolf–Rayet yields as well as including a fluorine production from novae, which could better model the observed data. Thus, the supersolar fluorine trend can currently not be explained theoretically.

Unfortunately, we do not have many stars in the metal-poor region. We have, however, one robust detection at $[\mathrm{Fe}/{\rm{H}}]=-0.87$, which shows a supersolar [F/Fe] value of +0.2 ± 0.1. This, together with our upper limits, and in light of the work by Guerço et al. (2019a) and Li et al. (2013), we cannot reject a (primary) contribution at early times. The AGB-star contribution is much too low at these metallicities. The models predicting a primary contribution at low metallicities are (i) the rapidly rotating, massive-star channel (Prantzos et al. 2018) and (ii) the ν process channel (most recently; Olive & Vangioni 2019). The trend from the rotating massive stars is a shallow and slightly larger-than-solar [F/Fe] for all metallicities (Prantzos et al. 2018), which is in better agreement with our data than the ν process prediction, which is lower for all metallicities.

The main s-process takes place in the interior of low- and intermediate-mass AGB stars (see, e.g., the discussion in Forsberg et al. 2019), and Ce is to 84% a main s-process element (Bisterzo et al. 2014). The [F/Ce] ratio should then be constant for the interval in metallicities where both elements are predominantly synthesized by the same type of AGB stars. We show the [F/Ce] ratio as a function of metallicity in Figure 10. Apart from the supersolar metallicities, the ratio is almost flat, although with some scatter. This strengthens the conclusion that AGB stars could be the dominant source of fluorine at least in the $-0.6\lt [\mathrm{Fe}/{\rm{H}}]\lt 0.0$ range.

Zoom In Zoom Out Reset image size

In the full sample of stars from which we have chosen our subset of stars, there are a number of stars that are enriched in s-process elements. These stars are significantly enriched in Zr, La, and Ce, simultaneously and by the same amount. The origin of these stars and the cause of this enrichment is not known, but being K giants, one possibility could be that the stars are in binary systems, i.e., they are extrinsic s-enhanced stars. The s-process element enhancement is then probably caused by pollution from the evolved AGB companion. This enhancement should then be accompanied by a fluorine enhancement, as fluorine is expected to correlate with the s-process elements, because both production chains include neutrons coming from He-nuclei reactions with ¹³C (Abia et al. 2019). In solar-metallicity carbon stars (on the AGB), this is indeed observed (Goriely & Mowlavi 2000; Abia et al. 2019). It would, therefore, be interesting to investigate whether these stars are enhanced in fluorine too.

We have determined upper limits of the fluorine abundances for two of these stars enhanced in s-process elements (2M14231899 and 2M17215666), both below $[\mathrm{Fe}/{\rm{H}}]\lt -0.8$, one being a thick disk and one a halo star. In Figures 2 and 3, we see that the upper limits for these stars are not especially high. On the contrary, they are even smaller than the robustly determined fluorine abundance from the star with a metallicity intermediate between the two stars. Thus, the fluorine abundances do not seem abnormally high, although we have very few data points in this metallicity range. The [F/Ce] ratio as a function of metallicity is plotted in Figure 10. The high Ce abundance and the normal fluorine abundances put these two stars below the general flat trend, and especially below the ratio of the star with the intermediate metallicity. We thus conclude that these s-enhanced stars do not show abnormally high fluorine abundances. This is surprising because it might have been expected had these elements originated from AGB stars.

We checked for binarity for these stars by looking for radial velocity shifts in spectra observed by APOGEE (SDSS IV; Majewski et al. 2017; Holtzman et al. 2018) at different epochs, but could not find any sign of binarity. Thus, the argument for them being extrinsic s-enhanced stars is weakened. Another possibility may be that they are globular-cluster escapees, which could leave traces of enrichment from AGB stars, and therefore the enhancement of the s-process elements. But then, fluorine enhancements could possibly also be expected.

It therefore seems unlikely that AGB stars, at these metallicities, are the cause of the s-process element enhancement or that fluorine is not produced at the same enhancement levels as the s-process elements from AGB stars. At these metallicities, the s-process elements could also have contributions from the r-process, which does not produce fluorine. The fluorine would then be produced in rapidly rotating massive stars or through the ν process.

We note that Cunha et al. (2003) also found that in their measurements of stars in ω Cen, stars with enhancements of s-process elements at similar metallicities, revealing a large contribution of AGB stars to the chemical evolution, did not show enhanced levels of fluorine either. These arguments strengthen the non-AGB contribution at lower metallicities. We also note, however, that Abia et al. (2019) conclude from analyzing carbon-rich AGB stars of different metallicities that models overpredict the production of fluorine for metal-poor stars ($[\mathrm{Fe}/{\rm{H}}]\lt -0.5$), which could potentially explain a lower F abundance than expected. It would be very interesting to measure the fluorine abundances in the other s-process-enriched stars from H. Jönsson et al. (2020, in preparation), which are at higher metallicities, in a range where the AGB-star contribution is high.

We can thus corroborate the results in Guerço et al. (2019a), who argue for a primary behavior at $[\mathrm{Fe}/{\rm{H}}]\lt -0.5$ and a secondary behavior above −0.5. This is supported by the fact that the s-enhanced stars in our sample do not show enhanced F for metallicities lower than $[\mathrm{Fe}/{\rm{H}}]\lt -0.6$.