Chapter 4
Modeling Cosmic Dust Evolution

 4.1 Stellar Evolution
  4.1.1 The Fate of Stars of Different Masses
   4.1.1.1 Nucleosynthesis
   4.1.1.2 Brief Outline of Stellar Evolution
   4.1.1.3 Parametrizing Star Formation
  4.1.2 Elemental and Dust Yields
   4.1.2.1 Injection of Heavy Elements in the ISM
   4.1.2.2 Production of Stardust
 4.2 Dust Evolution Processes in the ISM
  4.2.1 Grain Formation and Transformation
   4.2.1.1 Evidence of Grain Growth and Coagulation in the ISM
   4.2.1.2 Studies of the Magellanic Clouds
   4.2.1.3 Quantifying Grain Growth
  4.2.2 Grain Destruction
   4.2.2.1 Photodestruction of Small Grains
   4.2.2.2 Thermal Sputtering
   4.2.2.3 Destruction by SN Blast Waves
 4.3 Cosmic Dust Evolution
  4.3.1 Constraining the Dust Build-Up in Galaxies
   4.3.1.1 Cosmic Dust Evolution Models
   4.3.1.2 Empirical Inference of Dust Evolution Timescales
  4.3.2 Evolution of the Aromatic Feature Carriers
   4.3.2.1 The Different Evolution Scenarios
   4.3.2.2 The Observed Trends

This is not a new result – Draine & Salpeter (1979) reached the conclusion that grain destruction was rapid and that regrowth of dust in the ISM was required to explain the observed depletions. The numbers basically haven’t changed appreciably since then; the argument has been reiterated a number of times (...). Nevertheless, some authors continued to hold the view that the solids in the interstellar medium were primarily formed in stars.
 
(Bruce T. DRAINE;  Draine2009)

This chapter focusses on the study of dust evolution in all interstellar environments, at all spatial scales. Dust evolution is the variation of the constitution of a grain mixture with time, under the effects of its environment. The timescales of evolution being significantly longer than the career of a scientist, we usually study spatial variations of the dust content in a region, or the variations among a sample of galaxies. These different observations are then compared, being considered as snapshots at different evolutionary stages. The environmental parameters that are commonly used to quantify dust evolution are: (i) the ISM density and the ISRF intensity and hardness, for spatially-resolved studies; (ii) the metallicity and star formation rate, for global galactic studies. The main processes responsible for dust evolution are represented on Fig. 4.1.

Grain Formation
is the dust mass build-up by:
Grain Processing
is the alteration of the grain constitution in the ISM by:
Grain Destruction
is the full or partial removal of the elements constituting the grains by:
PIC
Figure 4.1: The interstellar dust lifecycle. This is the schematic representation of dust evolution through the ISM. The upper part illustrates the different sources, sinks and environments where grains are processed. The timescales are indicative and will be discussed in the rest of this chapter. The eight small bottom panels focus on the microscopic processes. Licensed under CC BY-SA 4.0.

4.1 Stellar Evolution

Stars have a crucial impact on ISD: (i) they synthesize the heavy elements that constitute dust grains (Fig. 2.16); (ii) they also directly produce dust seeds in their ejecta; (iii) the shock waves of SNe erode and vaporize the grains; (iv) the radiative and mechanical feedback of massive stars carve the ISM and process the grains.

4.1.1 The Fate of Stars of Different Masses

A star can be conceptualized as a sphere of gas in hydrostatic equilibrium, where the gravity is counterbalanced by the thermal pressure sustained by nuclear reactions in its core (e.g. Degl’Innocenti2016, for an introduction). The energy produced in the core is carried out through radiation, convection or conduction. The initial mass of a star, and in a lesser extent its initial metallicity, determine its future evolution.

4.1.1.1 Nucleosynthesis

The nuclear reactions in stellar interiors, on top of being the fuel of stars, lead to the production of heavy elements. A fraction of these freshly synthesized elements are injected back into the ISM, during the final stages of stellar evolution.

Nuclear binding energies. A fundamental quantity to determine the efficiency of nuclear reactions to sustain the thermal pressure within a star is the nuclear binding energy of an element of mass A (number of nucleons; cf. e.g. Chaps. 1-2 of  Pagel1997, for a review). This quantity is represented in Fig. 4.2 for the most relevant nuclei. To have an exothermic reaction, that will be able to counterbalance gravity, one needs to synthesize elements of higher binding energies. The curve of Fig. 4.2 reaches a maximum around 56Fe.

Fusion
of elements lighter than 56Fe is exothermic. Since the initial composition of a star is 34 H and 14 He, stellar nucleosynthesis takes this way.
Fission
of elements heavier than 56Fe is exothermic. This is the process implemented in nuclear reactors to generate electricity (through heat).
PIC
Figure 4.2: Average nuclear binding energies per nucleon. We have displayed the experimental data compiled by Ghahramany et al. (2012). The most stable elements are around 56Fe. Below, fusion is exothermic, above, it is fission. Licensed under CC BY-SA 4.0.

Primordial nucleosynthesis. Before the first stars appeared, 2H and 4He, as well as elements up to 7Li, were synthesized during the first 15 minutes after the big bang (e.g. Pagel1997Calura & Matteucci2004Johnson2019). The temperature was at this time around T 109 K. This primordial nucleosynthesis was brief, as the Universe was expanding and cooling. It is estimated that after 20 minutes, the temperature was too low to synthesize new elements. The primordial abundances refer to the elements produced during these first minutes (cf. Eq. 2.11):

Xprimordial 0.76,Y primordial 0.24,Zprimordial 0.00. (4.1)

Stellar nucleosynthesis. Once the temperature at the center of a collapsing protostar becomes high enough (T 107 K), thermonuclear reactions 1 are initiated. Several chains and cycles of reactions occur in stars, at different stages. The most important ones are the following (e.g. Filippone1986Pagel1997Silva Aguirre2018).

Proton fusion,
also called p-p chain, is a series of nuclear reactions converting 41H into 4He. This reaction chain has three branches cycling through various light elements (D, Li, B, Be). It is the dominant process in stellar interiors with T 2 × 107 K, that is for stars with mass m 1M. 2
CNO cycle
is another series of nuclear reactions converting 41H into 4He. Contrary to the p-p chain, this cycle requires pre-existing C, N or O (i.e. it requires a non-zero-metallicity star). This cycle can be broken into: (i) a CN cycle, starting with 12C and 41H, ending with 12C and 4He; and (ii) a NO cycle, starting with 15N and 31H, ending with 14N and 4He. This cycle is more efficient than the p-p chain for T 2 × 107 K, that is for stars with m 1M. In practice, both happen simultaneously, but with different efficiencies.
The triple α process
is a series of nuclear reactions converting 34He into 12C (the 4He nucleus is indeed called the α particle). It also produces 16O and 20Ne as byproducts. This process starts when the star has converted 10% of its H into He. It requires temperatures of T 108 K.
Heavier element fusion
happens essentially in massive stars (m 8M), when the temperature of the core reaches T 109 K. Several successive phases are then possible: C burning, Ne burning and O burning, producing up to 28Si. The last series of reactions are the α ladder, which produce elements up to Fe and Ni.

 H burning, which encompasses both the p-p chain and the CNO cycle, represents the longest phase in the lifetime of a star, whereas He burning lasts only 10% of its existence.

4.1.1.2 Brief Outline of Stellar Evolution

Stars are born from the collapse of molecular clouds into protostars (e.g. Motte et al.2018, for a review). Protostars accrete matter until their winds and radiation pressure stops this process, leading to a pre-main sequence star (pre-MS). Pre-MS stars exhibit violent winds and bipolar jets, clearing away the remaining molecular cocoon they were born in. They contract until the temperature in their core is high enough to initiate H fusion (T 107 K). Below m 0.08M, we get a brown dwarf , which is a compact object not massive enough to sustain nuclear reactions. Fig. 4.3 schematically represents the different stages of evolution of low- and high-mass stars.

PIC

Figure 4.3: Schematic representation of stellar evolution. Licensed under CC BY-SA 4.0.

The Main Sequence. Once nuclear reactions ignite, stars are on the Zero-Age Main Sequence (ZAMS; grey line in Fig. 3.18). We have already briefly discussed stellar evolution in Sect. 3.1.2.3. The different types of stars, their mass, luminosities and lifetimes are given in Fig. 3.18 and in Table 3.2.

1.
As long as stars are in their H burning phase (cf. Sect. 4.1.1.1) they are Main Sequence (MS) stars. They move slowly along their tracks in Fig. 3.18.
2.
Once the core has exhausted its H, it contracts by lack of fuel. This contraction increases the temperature, allowing the He burning to start (cf. Sect. 4.1.1.1). Due to the increase of central temperature, the outer layers expand. The star is now a red giant.
3.
This process repeats, with He burning (triple α process; Sect. 4.1.1.1).

The late stages of massive stars. Massive stars (8M M < 120M) are the hottest and most luminous ones (cf. Table 3.2). They are short-lived (τ(m) 30 Myr; Fig. 3.18).

1.
The process of burning heavier elements is repeated beyond C, resulting in an “onionstructure (cf. Fig. 4.3.a). The combustion of each element is exponentially faster. These stars are, at this point, red supergiants (cf. Fig. 3.18).
2.
Once the core is made of Fe, the star can not anymore produce energy by nuclear fusion. It therefore collapses (e.g. Heger et al.2003).
If m 40M,
the collapse is halted by the degeneracy pressure 3 of neutrons. The outer layers of the stars then explode as a type II supernova (SN II) or core-collapse supernova, leaving a Neutron Star (NS) in the center.
If m 40M,
the degeneracy pressure of the neutron core is not sufficient to sustain the collapse. The remnant is not anymore a NS, but a Black Hole (BH). This is also the approximate mass range where the star leaves a remnant without exploding as a SN, ending as a collapsar (Heger et al.2003). These two phenomena (ending as a collapsar and leaving a BH) are not necessarily concomitant. The exact masses above which these two phenomema occur are not accurately known and depend on other parameters, such as stellar rotation. For simplicity, we have represented both phenomena on the same branch in Fig. 4.3.

The late stages of LIMS. Low- and Intermediate-Mass Stars (LIMS; 0.08M M < 8M) are less luminous than massive stars, but they are the most numerous. Their lower gravity allow them to burn their elements slower than massive stars, and therefore to live longer (several Gyrs, on average; Fig. 3.18).

1.
Their mass does not allow them to start C burning. LIMS enter the Asymptotic Giant Branch (AGB; cf. Fig. 3.18; e.g. van Winckel2003Herwig2003, for reviews). They are larger and more luminous than red giants and are thermally pulsing.
2.
The contraction of the core is halted by electron degeneracy pressure. The thermal pulses lead the outer shell to expand progressively, creating a Planetary Nebula (PN), leaving a White Dwarf (WD) in the center.
3.
The maximum mass a white dwarf can reach is the Chandrasekar mass, mChandra 1.4M. If a white dwarf of mass mWD happens to be in a binary system with another red giant, it will accrete some of its mass.
If mWD mChandra,
the excess of mass above the Chandrasekar limit re-ignites the thermonuclear reactions. It follows a type Ia SN (SN Ia), which disrupts the binary system.
If mWD mChandra,
thermonuclear reactions are ignited at the surface of the white dwarf. It ensues a nova, that does not disrupt the binary system.

4.1.1.3 Parametrizing Star Formation

Star formation is a complex process involving stars of different masses being formed at different times. At the scale of a star-forming region or an entire galaxy, SF can be described statistically.

Initial mass functions. Initial Mass Functions (IMF) express the number distribution of stars of mass m born at a given time: ϕ(m) dNdm. IMFs are usually expressed in M1, and are normalized as 4:

mm+ ϕ(m)dm = 1, (4.2)

where m = 0.1M and m+ = 100M are the lower and upper masses. The average stellar mass is defined as:

mmm+ mϕ(m)dm. (4.3)

The fraction of stars ending their life as a core-collapse SN is:

fSN mSNm+SN ϕ(m)dm, (4.4)

with mSN = 8M and m+SN = 40M (e.g. Heger et al.2003). Several IMFs have been proposed in the literature (see also  Kroupa2001).

The Salpeter IMF
(Salpeter1955) was the first one proposed. It is defined as:
ϕSalp(m) (1 α) × mα m+1α m1α, with α = 2.35 (4.5)

where the lower and upper masses, m and m+, are usually taken as m = 0.1M and m+ = 100M, although the original Salpeter (1955) study constrained the index of the power-law only up to m+ = 10M.

The Chabrier IMF
(Chabrier2003) for individual stars is defined, within the same mass range, as:
ϕChab(m) 0.16033 × m2.3  for m > 1M 0.5718 m exp (log 10(m) + 1.1023729)2 0.9522  for m 1M. (4.6)
The top-heavy IMF
(e.g. Dwek et al.2007) is often invoked at high-redshift. It is defined as:
ϕTop(m) (1 α) × mα m+1α m1α, with α = 1.50 (4.7)

These IMFs are compared in Fig. 4.4. Some of their properties are listed in Table 4.1. The IMF is thought to be a universal property of interstellar media. The different IMFs of Table 4.1 have consequences on the stellar properties. The current consensus is that the Chabrier IMF might be more appropriate than Salpeter’s, at least at low redshift.

PIC PIC
Figure 4.4: Initial mass functions. We compare the three IMFs discussed in the text: (i)  Salpeter (1955) (Eq. 4.5); (ii) Chabrier (2003) (Eq. 4.6); and (iii) Top-heavy (Eq. 4.7). Panel (a) shows the number distribution of stars, mϕ(m). Panel (b) shows the luminosity distribution, Lmϕ(m), where L comes from Fig. 3.18. We have indicated in yellow the range 8 m < 40M, corresponding to SN II. Licensed under CC BY-SA 4.0.





Salpeter Chabrier Top-Heavy




Average mass, m

0.351M 0.673M 3.16M




SN II fraction, fSN

0.239% 0.724% 6.38%




Mass fraction of massive stars

13.9% 22.6% 74.1%




LIMS luminosity (t = 0, Z = 0.008)

4.81LM 13.9LM 55.3LM




Massive star luminosity (t = 0, Z = 0.008)

180LM 579LM 1.38 × 104L M




Total luminosity (t = 0, Z = 0.008)

185LM 593LM 1.38 × 104L M




Table 4.1: IMF properties. These properties are integrated over three different IMFs: Salpeter (Eq. 4.5); Chabrier (Eq. 4.6); and Top-heavy (Eq. 4.7). For the last three lines, we used the ZAMS stellar luminosities of Fig. 3.18 (initial metallicity Z = 0.008).

Parametric star formation histories. We have already briefly discussed Star Formation Histories (SFH) in Sect. 3.1.2.3. The SFH quantifies the SFR, ψ(t), as a function of time, t, of a star-forming region or galaxy. Several parametric forms are commonly used in the literature. As we will see in Sect. 4.3, their parameters can be inferred by fitting a set of observations.

The exponential SFH
(cf. Fig. 4.5.a) is parametrized by a timescale, τSF, and a SFR at t = 0, ψ0:
ψexp(t) ψ0 exp t τSF . (4.8)
The delayed SFH
(cf. Fig. 4.5.bLee et al.2010) has the same number of parameters as the exponential SFH, but its SFR peak is delayed at t = τSF:
ψdel(t) ψ0 t τSF exp t τSF . (4.9)

It is possible to combine several SFHs to account for the complex history of a galaxy. A useful quantity, deriving from the SFH, is the stellar birth rate, which is the average number of stars born per unit time:

B(t) ψ(t) m. (4.10)

The rate of SN II, RSN(t), can be approximated from this quantity:

RSN(t) mSNm+SN B t τ(m) × ϕ(m)dm B(t)fSN, (4.11)

where τ(m) is the lifetime of a star of mass m (cf. Fig. 3.18). The approximation, in the second part of Eq. (4.11), comes from the fact that the lifetime of massive stars (τ(m) 30 Myr) is usually much smaller than the SF timescale: τ(m) « τSF.

PIC
Figure 4.5: Parametric star formation histories. Panel (a) represents the exponential SFH (Eq. 4.8), for different values of τSF. Panel (b) represents the delayed SFH (Eq. 4.9), for the same grid of τSF. In both panels, the SFHs are normalized so that the integrated mass of stars formed in Δt = 14 Gyr (about the age of the Universe) is M = 4 × 1010M (roughly the MW stellar mass). Licensed under CC BY-SA 4.0.

4.1.2 Elemental and Dust Yields

Stars, in their late stages, return to the ISM a fraction of the heavy elements they have synthesized. Stellar ejecta are: (i) stellar winds; (ii) planetary nebulae; (iii) novae; (iv) SN Ia; and (v)  SN II. In addition, the temperature in these ejecta can be low enough (cf. Fig. 2.17.b) to condense grains, that we refer to as stardust.

4.1.2.1 Injection of Heavy Elements in the ISM

Fig. 4.6 gives the approximate fraction of each element produced in different environments, for the Solar neighborhood (e.g. Johnson2019). These proportions depend on the past SFH of the system we are considering.

Primordial nucleosynthesis
(cf. Sect. 4.1.1.1) is responsible for the production of most of the light elements.
SN II
account for most of the ISD-relevant elements, except C.
AGB ejecta
(i.e. LIMS winds and PNe) account for a significant fraction of C and N, and heavier elements we have not displayed here.
SN Ia
are responsible for synthesizing a significant fraction of the metals around Fe.
Other processes,
such as cosmic ray fission and merging neutron stars, are not very relevant to ISD.
PIC
Figure 4.6: Nucleosynthesis origin of the main elements. This curve represents the Solar abundances of Asplund et al. (2009) that we have shown in Fig. 2.16. We have indicated the fraction of the different sources of nucleosynthesis where these elements originate from (Table S1 of  Johnson2019). Since we have stopped at Z = 36, none of the elements shown are coming from merging neutron stars. Licensed under CC BY-SA 4.0.

Stellar elemental yields. A stellar yield, Y E(m), is the mass of an element E injected into the ISM by a star of mass m, at the end of its lifetime. These yields can be constrained observationally, but they are essentially determined theoretically (e.g. Karakas & Lattanzio2014, for a review). Fig. 4.7.a shows the yields of the most important elements. An important quantity determining the type of dust grains that will form in the ejecta is the C/O ratio. Indeed, when the temperature cools down enough, C and O tend to combine to form CO molecules. The excess atom will thus be the only one left to form stardust. Therefore, stellar ejecta with C < O will form primarily O-rich grains (silicates and oxides), whereas stellar ejecta with C > O will form mainly carbon grains and SiC. Fig. 4.7.b compares the number abundances of C and O ejected by stars of different masses. We can see that carbon grains originate mainly in LIMS around m 3M.

 SN II are responsible for most O-rich stardust, while LIMS produce most C-rich grain seeds.

PIC PIC
Figure 4.7: Solar metallicity elemental stellar yields. Panel (a) shows the mass of the most abundant elements ejected by stars of different masses. Panel (b) focusses on the stellar mass range where the number abundance of C exceeds that of O (in yellow). These data are the yields used by the chemical evolution model of Galliano et al. (2008a). They are an homogenization of the yields of Karakas & Lattanzio (2003a,b), for LIMS, and of Woosley & Weaver (1995), for high-mass stars. Licensed under CC BY-SA 4.0.

Metallicity estimates. The heavy elements injected by the successive stellar populations increase the metallicity of the ISM, Z. In that sense, the metallicity is an indicator of “the fraction of baryonic matter that has been converted into heavier elements by means of stellar nucleosynthesis” (Kunth & Östlin2000), such that:

Xprimordial 0.76 Y primordial 0.24 Zprimordial 0.00 X 0.74 Y 0.25 Z 0.01. (4.12)

The measure of metallicity is not completely straightforward and is vigorously debated (e.g. Kewley et al.2019, for a review). In external galaxies, it is usually estimated by modeling observations of nebular optical lines, coming from H II regions. Several methods, allowing an observer to convert a few line ratios into a metallicity estimate, have been proposed. These methods have been calibrated on particular H II regions, modeling the photoionization, making unavoidable assumptions about the stellar populations and the topology of the gas. We have systematically compared several calibrations over the DustPedia sample (De Vis et al.2019). We have favored the “S” calibration from Pilyugin & Grebel (2016), as it is the most reliable down to low metallicities.

4.1.2.2 Production of Stardust

At the scale of a galaxy, the two most important sources of stardust are: (i) AGB stars, encompassing both LIMS winds and PNe; and (ii) SN II.

AGB stars. Most of the dust production in LIMS is believed to occur during the Thermally-Pulsing Asymptotic Giant Branch (TPAGB) phase (Gail et al.2009). In addition, LIMS with m 1M do not condense grains (e.g. Ferrarotti & Gail2006). Theoretical models concur that only a fraction of the available heavy elements will go into stardust (Morgan & Edmunds2003Ventura et al.2012):

δLIMS mstardustej mZej 10 40%, (4.13)

mstardustej and mZej being the ejected mass of stardust and heavy elements. Observations and modeling of the circumstellar envelopes of post-AGB stars are consistent with these values (e.g. Ladjal et al.2010).

SN II. There is solid evidence that grains form in SuperNova Remnants (SNR), as the ejected gas cools down. Theoretical estimates of the net dust yield of a single SN II range in the literature from Y SN 103 to Y SN 1MSN (e.g. Todini & Ferrara2001Ercolano et al.2007Bianchi & Schneider2007Bocchio et al.2016Marassi et al.2019). From an observational point of view, measuring the dust mass produced in situ by a single SN II is quite difficult, as it implies disentangling the freshly-formed dust from the surrounding ISM. It also carries the usual uncertainty about dust optical properties. A decade ago, the largest dust yield ever measured was Y SN 0.02M (in SN2003gd;  Sugerman et al.2006). The Herschel space telescope has been instrumental in estimating the cold mass of SNRs. The yields of the three most well-studied SNRs are now an order of magnitude higher:

Cassiopeia A:
Y SN 0.04 1.1M (Barlow et al.2010Arendt et al.2014De Looze et al.2017Bevan et al.2017Priestley et al.2019);
The Crab nebula:
Y SN 0.03 0.23M (Gomez et al.2012Temim & Dwek2013De Looze et al.2019);
SN 1987A:
Y SN 0.45 0.8M (Dwek & Arendt2015Matsuura et al.2015).

Most of the controversy however lies in the fact that, while large amounts could form in SN II ejecta (e.g. Matsuura et al.2015Temim et al.2017), a large fraction of freshly formed grains would not survive the reverse shock (v 1000 km/s; e.g.  Nozawa et al.2006Micelotta et al.2016Kirchschlager et al.2019). In all the cases we have listed above, the newly-formed grains have indeed not yet experienced the reverse shock (Bocchio et al.2016). The net yield is thus expected to be significantly lower. Even if 10 20% of the dust condensed in an SN II ejecta survives its reverse shock (e.g. Nozawa et al.2006Micelotta et al.2016Bocchio et al.2016), we have to also consider the fact that massive stars are born in clusters. The freshly-formed dust injected by a particular SN II, having survived the reverse shock, will thus be exposed to the forward shock waves of nearby SNe (e.g. Martínez-González et al.2018). Overall, SN II dust yields are largely uncertain. We will extensively discuss their empirical constraint, from a statistical point of view, in Sect. 4.3. We will show that we can infer the average dust yield per SN II, Y SN. The corresponding timescale is then simply:

1 τSNcond(t) Y SN Mdust(t)RSN(t). (4.14)

Indirect evidence. The best constraints on the fraction of ISD which is stardust might be indirect. The clear correlation between the depletion factor, F, and the average density of the ISM, nH, that we have discussed in Fig. 2.17.a, has been shown to require rapid destruction and reformation into the ISM (e.g. Draine & Salpeter1979Tielens1998Draine2009). The rest of the grains needs to form in the ISM. Draine (2009, D09) gives a series of additional arguments concluding that, in the MW, stardust has to be less than 10% of ISD.

1.
A first argument given by D09 is based on the typical lifetime of ISD.
a)
From the literature, the stardust injection rate is roughly stardust 5 × 103M yr.
b)
Noting that the typical lifetime of a dust grain in the MW is τISD 3 × 108 yr (cf. Sect. 4.2.2.3;Jones et al.1996), the present stardust mass in the ISM should be: Mstardust stardust × τISD 1.5 × 106M .
c)
The ISD mass in the MW is roughly MISD MISM183 2.7 × 107M (cf. Table 2.4).

Stardust is thus only MstardustMISD 5% of the total ISD.

2.
A second argument given by D09 is based on the study of IDPs in meteorites (cf. Sect. 2.2.4).
a)
Stardust silicate grains in meteorites, identified with their isotopic anomalies (cf. Fig. 2.19), appear to be 20% crystalline.
b)
Silicates in the ISM are less than 2% crystalline (cf. Sect. 2.2.1.2Kemper et al.2004).

Therefore, the fraction of stardust is only less than 220 10% of ISD. Also, the fact that ISD is mainly amorphous, whereas circumstellar grains are essentially crystalline, is another argument in favor of rapid destruction and reformation in the ISM.

 In the MW, stardust represents only a few percents of the ISD content.

4.2 Dust Evolution Processes in the ISM

Most of the important dust evolution processes occur in the ISM. These effects can be studied by looking at spatial variations of the dust properties in a region.

4.2.1 Grain Formation and Transformation

Grain formation is the transfer of elements from the gas phase to the dust phase, therefore increasing the dustiness.

4.2.1.1 Evidence of Grain Growth and Coagulation in the ISM

We have just discussed stardust (cf. Sect. 4.1.2.2) which is thought to produce grain seeds onto which mantle can grow. We now focus on the dominant process in Solar metallicity systems: the accretion of gas phase atoms and molecules. Grain-grain coagulation does not result in grain formation per se, as it does not affect the dustiness. It however follows grain growth and has similar effects on the FIR opacity.

The evidence brought by depletions. As we have discussed in Sect. 4.1.2.2, the clearest evidence of grain growth in the ISM is provided by the good correlation between the depletion factor and the average density of the ISM (cf. Fig. 2.17.a). It implies that atoms and molecules from the gas phase are progressively building up grain mantles, when going into denser regions. This observed behaviour is also consistent with the progressive de-mantling and disaggregation of cloud-formed, mantled and coagulated grains injected into the low density ISM, following cloud disruption. It is perhaps not unreasonable to hypothesise that dust growth in the ISM occurs on short timescales during cloud collapse rather than by dust growth in the quiescent diffuse ISM. In this alternative interpretation, the arrow of time is in the opposite sense and requires rapid dust growth, through accretion and coagulation, in dense molecular regions and slow de-mantling and disaggregation in the diffuse ISM (e.g. Jones2009). Given that astronomical observations provide only single-time snapshots, it will seemingly be difficult to determine the direction of the time-arrow of dust evolution.

FIR opacity variations. We have seen in Fig. 1.21 that the growth of mantles has an impact on the FIR opacity (cf. Köhler et al.20142015). Yet, there is clear evidence of FIR opacity variations in the MW. The main factor seems to be the density of the medium. For instance, both Stepnik et al. (2003) and Roy et al. (2013) found that the FIR dust cross-section per H atom increases by a factor of 3 from the diffuse ISM to the molecular cloud they targeted. Stepnik et al. (2003) noticed that this opacity variation is accompanied by the disappearance of the small grain emission. They concluded that grain coagulation could explain these variations. In the diffuse ISM, Ysard et al. (2015) showed that the variation of emissivity, including the β T relation (cf. Sect. 3.1.2.1), could be explained by slight variations of the mantle thickness of the THEMIS model. For that reason, the THEMIS model aims at describing the evolution of grain mantles as a function of density and ISRF, as we have seen in Fig. 1.21: (i) in the diffuse ISM, the grains are supposed to have a thin a-C mantle, largely dehydrogenated (aromatic) by UV photons; (ii) in denser regions, the mantle thickness is hypothesized to increase and to become more hydrogenated (aliphatic), because of the progressive shielding of stellar photons; (iii) in molecular clouds, grains are thought to be coagulated and iced. The THEMIS model predicts a factor of 2.5 dust mass increase in going from MW diffuse to dense clouds. This would correspond to a factor up to 7 in terms of dust emissivity per H atom (Köhler et al.2015).

4.2.1.2 Studies of the Magellanic Clouds

In nearby galaxies, studies of the local grain processing are difficult to conduct, as the emissivity variations are smoothed out by the mixing of dense and diffuse regions. Even when potential evolutionary trends are observed, their interpretation is often degenerate with other factors. The Magellanic clouds are the most obvious systems where this type of study can be attempted. The insights provided by depletion studies (cf. Sect. 2.2.3) show that there are clear variations of the fraction of heavy elements locked-up in dust, and these variations correlate with the density (Tchernyshyov et al.2015Jenkins & Wallerstein2017). Since the coagulation and the accretion of mantles lead to an increase of FIR emissivity (cf. Fig. 1.21Köhler et al.2015), we should expect emissivity variations in the Magellanic clouds. Indeed, Roman-Duval et al. (2017) studied the trends of gas surface density (derived from H I and CO) as a function of dust surface density (derived from the IR emission), in these galaxies. They found that the observed dustiness of the LMC increases smoothly by a factor of 3 from the diffuse to the dense regions. In the SMC, the same variation occurs, with a factor of 7. They argue that optically thick H I and CO-free H2 gas (cf. Sect. 3.3.2.3) can not explain these trends, and that grain growth is thus the most likely explanation.

Spatially-resolved SED fitting of LMC-N44 and SMC-N66. We have conducted a similar study, focussing on two massive star-forming regions, rather than the whole galaxies 5: (i)  N44 in the LMC; and (ii) N66 in the SMC (Galliano2017). Our maps were 200 pc wide regions, with a spatial resolution of 15 pc. The MIR-to-submm data were coming from the SAGE/HERITAGE surveys (Spitzer and Herschel data; Meixner et al.20062013). We used the hierarchical Bayesian SED model, HerBIE, with the AC dust composition of Galliano et al. (2011, cf. Sect. 3.1.3.3). The goal was to perform a spatially-resolved modeling of the dust properties, in a region with a strong gradient of physical conditions, in order to probe dust processing, as a function of density, ISRF and metallicity. The wide range of physical conditions can be estimated by looking at the range of SEDs shown in Fig. 4.8. In both panels, the faintest pixels show a rather cold SED, peaking around λ 100μm, whereas the brightest pixels peak around λ 60μm, with a very broad FIR bump, indicating a wide range of ISRF intensities, typical of compact SF regions (cf. Sect. 3.1.2.2).

PIC

Figure 4.8: Spatially-resolved SED fits in N44 and N66. In both panels, we show the brightest and the faintest pixels of our entire map. The grey error bars are the observations, and the colored densities are the Bayesian posterior distributions of SED models (Galliano2017). Licensed under CC BY-SA 4.0.

Derived dust-gas relations. We have compared the derived dust and total gas column densities. The latter was estimated from the H Iline and 12CO(J =1 0) 2.6mm measurements (Meixner et al.2006Gordon et al.2011Meixner et al.2013). The results are displayed in Fig. 4.9. The orange line represents the Galactic dustiness (cf. Table 2.4) scaled by the metallicity, therefore representing the Galactic dust-to-metal mass ratio. This line corresponds to the values we would expect if the dust constitution was close to the diffuse ISM of the MW and was not evolving with density. The most diffuse pixels in both regions are consistent with this value. The hatched yellow area corresponds to a dustiness larger than the metallicity, that is requiring more heavy elements in dust than what is available in the ISM. Overall, the trends of Fig. 4.9 indicate a non-linear dust-to-gas relation, with a variation of the observed dustiness by a factor of 3, similar to the studies we have reviewed at the beginning of this section. The high density pixels lie in the forbidden zone. The possible causes are the following.

Grain growth:
a part of this trend is likely the result of the evolution of the true dustiness, due to mantle accretion in denser regions. The yellow hatched area can however be considered as a hard upper limit, as in practice, not all heavy elements are refractory: even in the densest molecular clouds, there are gas phase CO, HCN, etc. This sole factor is therefore probably not sufficient to explain the full extent of the observed dustiness variation.
Emissivity increase:
grain growth and coagulation are accompanied by an increase of FIR emissivity (cf. Fig. 1.21Köhler et al.2015). This effect is naturally expected and would amplify the increase of the observed dustiness with density, as the constant emissivity assumption of our SED model would lead us to overestimate the dust mass of dense regions.
Contribution of dark gas:
we have not accounted for CO-dark gas (cf. Sect. 3.3.2.3). This component can potentially bias the molecular gas mass estimate in translucent regions by up to a factor 100 (Madden et al.2020). It could also explain a part of the trend, as it would result in an underestimate of the total gas mass. This effect should however be significative at intermediate column densities, and decrease toward the densest regions, where CO would dominate. It might thus not be the main cause of the non-linearity of our trends.

 There is multiple evidence of dust evolution as a function of density, consistent with grain growth and coagulation, and the consequent increase of emissivity.

PIC
Figure 4.9: Dust-to-gas mass surface density relation in N44 and N66. In both panels, we show the trends of dust mass surface density as a a function of the total gas mass (atomic and molecular) surface density derived from the SED fitting of Fig. 4.8 (Galliano2017). Each point, with its uncertainty ellipse, corresponds to 15 pc pixel. The orange line corresponds to the dustiness of the MW (cf. Table 2.4) scaled by the metallicity of the region. The yellow hatched area corresponds to a dustiness larger than the metallicity, that is an unrealistic value requiring more heavy elements being locked up in grains than what is available in the ISM. Licensed under CC BY-SA 4.0.

4.2.1.3 Quantifying Grain Growth

We now discuss the way grain growth can be approximately quantified. The following relations are rather uncertain, because of the lack of constraint on grain structure and composition. They however provide a framework to study grain growth efficiency.

Accretion timescale. Timescales for grains to accrete atoms are widely discussed in the literature (e.g. Dwek1998Edmunds2001Draine2009Hirashita & Kuo2011Zhukovska et al.2016Priestley et al.2021). First, the collision rate of an atom E of mass mE, with a grain of radius a is:

1 τcoll(a,E) πa2 grain cross-section ×nE gas density of E ×8kT gas πmE Maxwellian velocity of E. (4.15)

In this equation, we have implicitly neglected Coulomb interaction (i.e. we have assumed that the grain and the atom are both neutral, which is a reasonable assumption in the CNM). Second, the growth rate of a grain of mass md(a), due to accretion following these collisions, can be written:

dmd dt acc(a,E) = Ssticking probability × mE fE gained mass × 1 τcoll(a,E) rate, (4.16)

where 0 S 1 is the sticking coefficient, that is the probability the atom will be bound with the grain after the collision. The factor fE is the mass fraction of element E within the grain. We choose E as a key element (Zhukovska et al.2008), that is the element in the grain make-up that will have the longest collision time.

For silicates,
the key element is Si, with fSi 0.16 for olivine and fSi 0.24 for pyroxene (assuming Fe:Mg=1:1). In other words, for each collision with Si, there are more collisions with O, Fe and Mg. Therefore, the dust mass gained between two Si collisions is the mass of a full crystal unit (SiO4MgFe for olivine SiO3Mg0.5Fe0.5 for pyroxene; cf. Sect. 1.1.4.1).
For carbon grains,
the mass is essentially C, as H is negligible. We thus have fC 1.

Finally, it is convenient to express this quantity as an accretion timescale, τacc(a):

1 τacc(a) 1 md(a) dmd dt acc(a,E) = SmE fE 3nE 22kT gas π , (4.17)

where we have simply developed md(a) = 43πa3ρ in the second equality, ρ being the mass density of the grain. The density of the element E can be written as a function of the total H density, assuming its abundance scales with metallicity:

nE Z Z E HnH. (4.18)

We therefore see that the grain growth timescale roughly obeys the following proportionality (assuming S = 1 and olivine composition of silicates):

τacc(a) 100cm3 nH ×Z Z ×100K Tgas × a 100nm 57Myr for silicates 41Myr for carbon grains. (4.19)

As said above, these estimates are uncertain. We especially have no idea of the sticking probability, S. Eq. (4.19) however provides a description of the sensitivity of grain growth to density, size and metallicity. It is also indicative of the lower limit of these timescales.

Grain growth in different ISM phases. Fig. 4.10 displays Eq. (4.19) for carbon and silicate grains in the most relevant ISM phases. Timescales longer than the typical destruction timescales by SN II blast waves are irrelevant. That is the reason why this range is hatched in yellow in Fig. 4.10.

In the WNM,
only the smallest grains (a 10 nm) could grow. The WNM is thus not very suitable for grain growth.
In the CNM,
all grain sizes can grow in less than 300 Myr. The CNM is thus a phase where dust growth could happen.
In diffuse molecular clouds,
growth timescales are roughly similar to the CNM. The same conclusion therefore applies.
In dense molecular clouds,
all relevant interstellar grain sizes can grow in less than 10 Myr, which is also the typical lifetime of these clouds, in star-forming regions.

These timescales are consistent with the picture painted by the variation of elemental depletions across phases (cf. Fig. 2.17.a). To estimate a global growth timescale, let’s consider the radius corresponding to the average mass of the THEMIS model, in Table 2.3:

a3 a3 31nm for silicates 28nm for large a-C(:H). (4.20)

With these sizes, a typical accretion time in the CNM would be τacc 58 Myr for silicates, and τacc 38 Myr for large a-C(:H).

 Grains can possibly grow in the CNM, on timescales of 30 60 Myr, and faster in molecular clouds.

PIC
Figure 4.10: Grain growth timescales. Both panels represent Eq. (4.19) for several of the phases in Table 3.6 (taking nH = 104 cm3 for the dense molecular phase). We have assumed S = 1 for both silicates and carbon grains. We have highlighted timescales longer than the typical shock destruction timescale, τSNdest 300 Myr (cf. Sect. 4.2.2.3), in hatched yellow. Licensed under CC BY-SA 4.0.

Relation to global parameters. As we will see in Sect. 4.3, it is convenient to relate the grain growth timescale to global galaxy parameters. Mattsson et al. (2012) proposed a relation based on the following assumptions.

1.
Most of grain growth happens in molecular clouds. The mass surface density of these molecular clouds, Σmol, is proportional to the SFR surface density, ΣSFR (e.g. Kennicutt1998b). Grain growth rate is thus proportional to ΣSFR:
Σgas τgrow ΣSFR. (4.21)
2.
The grain growth rate is also proportional to the fraction of available heavy elements in the gas. It implies that:
Σgas τgrow ΣSFR × 1 Zdust Z Z, (4.22)

where Zdust is the dustiness, and ZdustZ, the dust-to-metal mass ratio (cf. Sect. 2.2.3.2). By subtracting ZdustZ, we account for the fact that the fraction of heavy elements already locked up in grains does not contribute to grain growth.

3.
The other parameters in Eq. (4.19) are assumed to not vary significantly. These parameters are the mean grain size, the mean gas velocity and mean density of molecular clouds.

The grain growth rate proposed by Mattsson et al. (2012) can thus be parametrized as a function of global galactic quantities and a phenomenological, dimensionless parameter, 𝜖grow, containing all our uncertainties. The goal is to empirically infer 𝜖grow, as we will see in Sect. 4.3. Eq. (4.22) thus becomes:

1 τgrow(t) = 𝜖grow ψ(t) Mgas(t)(Z(t) Zdust(t)), (4.23)

where we have replaced the ratio of surface densities by the ratio of the quantities, and have explicited the temporal dependencies. In the case of the MW (ψ 1.3M/yr; Mgas 7 × 109M ), a grain growth timescale of τgrow 60 Myr (Eq. 4.19) corresponds to 𝜖grow 104.

4.2.2 Grain Destruction

We now discuss grain destruction, that is the return of heavy elements from the grains to the gas phase. Note that fragmentation and shattering by shock waves (at v 200 km/s), that we have discussed in Sect. 4.2.2.3, simply rearrange the size distribution without destroying the dust. Shocks however have a pulverization effect, accompanying the other processes, that are difficult to differentiate from an observational point of view.

4.2.2.1 Photodestruction of Small Grains

Due to thermal spikes, small grains have a certain probability that one of their atom will be ejected. This is a runaway process leading to the complete sublimation of the dust grain.

Photodesorption and sublimation. Following the formalism of Guhathakurta & Draine (1989), we consider a cluster XN containing N atoms of X (X can be C, Fe, Si, O, etc.). The ejection of an atom from the grain is balanced by the return of an atom from the gas phase. The rate of the reaction XN + X XN+1 is RNAN, where the total grain surface is AN = 4σN 4πa2. Guhathakurta & Draine (1989) write the sublimation rate as:

dN dt = RN(T) × SN(T) × AN, (4.24)

and provide the following rates for graphite and silicate:

RN+1gra(T) 4.6 × 1033 αN 0.1 exp Bgra kT m2s1, RN+1sil(T) 4.9 × 1034 αN 0.1 exp Bsil kT m2s1, (4.25)

with the binding energy per atom:

Bgrak = 81200 20000N13K, Bsilk = 68100 20000N13K. (4.26)

The sticking coefficients, αN, is unknown and is arbitrarily chosen by the authors to be αN 0.1. Assuming the surface free energy is about 2 × 104 K, the term 20000N13 accounts for the surface tension, making it easier to release an atom when the grain is smaller. Finally, the suppression factor, SN(T) < 1, accounts for the suppression of the thermal fluctuations in a thermally isolated particle. This factor is:

SN(T) = 1 + γ γ bΓ(γf + 1)Γ(γf + f b) Γ(γf b + 1)Γ(γf + f), (4.27)

where f = 3N 6 is the number of vibrational degrees of freedom (cf. Sect. 1.2.3). In addition, the mean number of quanta per degree of freedom is γ = H(T)(ω0f), and the number of quanta necessary to release a particle is b = B(ω0). Guhathakurta & Draine (1989) take ω0 = 0.75Θ, where Θ is the Debye temperature (taking Θgra = 420K and Θsil = 470K; cf. Sect. 1.2.3.2). The mean lifetime is then integrated over the temperature distribution:

1 τsubl =dP dTRN(T)SN(T)ANdT. (4.28)

Guhathakurta & Draine (1989) assume that a grain does not survive if it has a lifetime 1013s 0.3 Myr. Fig. 4.11 displays these lifetimes for silicates and graphite bathed in the Mathis et al. (1983) ISRF. Although the exact numbers are to be taken with caution, we can conclude the following.

1.
In the diffuse ISM of the MW (U = 1), silicates larger than a 4.5 Å and graphite a 3.5 Å can survive.
2.
When a increases, the lifetimes become exponentially longer, meaning that for grains larger than a 6 Å, the survival of these grains will not be very sensitive to the assumptions in Eq. (4.28).

The hardness of the ISRF, that we have not represented here, will however increase the minimum size a grain needs to have in order to survive. The vicinity of OB associations will thus be environments where the smallest grains can be photodestroyed.

PIC
Figure 4.11: Lifetimes of small grains in a radiation field. In both panels, we plot the mean sublimation times from Eq. (4.28), for different grain radii (color lines), as a function of the starlight intensity of the Mathis et al. (1983) ISRF: (a) for the silicates of Weingartner & Draine (2001a); and (b) for the graphite of Laor & Draine (1993). Licensed under CC BY-SA 4.0.

Evidence in resolved regions. This last point is observationally verified in countless regions. It can be conveniently witnessed, as the smallest grains are the carriers of the MIR continuum, which is well separated from the rest of the emission (cf. Fig. 2.27.b). In addition, small carbon grains carry the series of aromatic features (cf. Sect. 3.2.1.1). The disappearance of these features in regions of enhanced ISRF is very likely the sign of the destruction of these grains by hard UV photons. This is, for instance, evident in one of our studies of the massive star-forming region, N11, in the LMC (Galametz et al.2016). This region contains several blobs, with embedded star clusters. The maps of the PAH mass fraction, qPAH (cf. Sect. 3.1.2.2), is shown in Fig. 4.12.a. It has been derived by modeling the spatially-resolved SED of Spitzer and Herschel images. Comparing this image to the mean starlight intensity in Fig. 4.12.b, we see that PAHs are strongly depleted in the blobs where U is enhanced. The photodestruction is evident. In the case of this massive region, a bright star cluster such as N11B can clear PAHs out over a region of typically 50 pc.

 Aromatic features are severely depleted around star-forming regions.

PIC PIC
Figure 4.12: Carving out of PAHs by UV photons in N11. Both maps have identical fields of view, centered on the star cluster N11B (LH10;  Lucke & Hodge1970), and are 400 pc wide. The quantities displayed are the results from the spatially-resolved SED modeling, with the composite approach (cf. Sect. 3.1.2.2;Galametz et al.2016): (a) the PAH mass fraction, qPAH; and (b) the mean starlight intensity, U, in units of 2.2 × 105 W/m2. The black areas have been masked. Licensed under CC BY-SA 4.0.

4.2.2.2 Thermal Sputtering

Another destruction mechanism is grain erosion and vaporization by collisions with energetic ions, either in coronal plasmas (thermal sputtering; T 106 K), or in shock waves (kinetic sputtering; v 100 km/s). There is an abundant literature on the subject (e.g. Draine & Salpeter1979Dwek & Scalo1980Tielens et al.1994Jones et al.1996Jones2004Nozawa et al.2006Micelotta et al.2010Bocchio et al.20122014Hu et al.2019, see also the review by Dwek & Arendt, 1992). We start by discussing thermal sputtering in this section, and will review kinetic sputtering in Sect. 4.2.2.3.

Sputtering times. The evolution of a grain of radius, a, and mass, md(a) = 43πa3ρ, subjected to sputtering in a gas of density nH, can be expressed (e.g. Hu et al.2019):

dmd(a) dt = 3md(a)1 a da dt = 3md(a) a nHY sput(Tgas,vs), (4.29)

where we have hidden all the microphysics into the sputtering yield, Y sput(Tgas,vs) dadtnH. This quantity depends on: (i) the gas temperature, Tgas, in case of thermal sputtering; or (ii) the shock velocity, vs, in case of kinetic sputtering. A detailed derivation of Y sput can be found in Nozawa et al. (2006, Sect. 5). For our simple discussion, we will adopt their yields, for silicate and carbon grains, fitted by Hu et al. (2019). In the thermal case, the sputtering rate can be expressed as:

1 τsputth(a,Tgas,nH) 1 md dmd dt = 3nH a Y sput(Tgas). (4.30)

Fig. 4.13.a shows the lifetimes of grains in a coronal plasma. With Eq. (4.30), in the HIM (cf. Table 3.6), typical grains (Eq. 4.20) have lifetimes of: (i) τsputth 11 Myr, for silicates; and (ii) τsputth 16 Myr, for carbon grains. In the case of SN II blastwaves, grains stay in post-shock conditions for only 104 yr. Dust destruction by thermal sputtering is thus not the dominant process in the shocked ISM.

 Grains have short lifetimes in coronal plasmas.

PIC
Figure 4.13: Thermal and kinetic sputtering times of silicates and carbon grains. Panel (a) show the mean lifetimes of grains in a hot gas, computed from Eq. (4.30), using the sputtering yields of Nozawa et al. (2006) fitted by Hu et al. (2019). Panel (b) show the mean lifetimes of grains in a shock of velocity vs, computed from Eq. (4.31), using the same sputtering yields. Licensed under CC BY-SA 4.0.

Early-type galaxies. We have seen in Sect. 3.1.3.1 that ETGs tend to be characterized by a diffuse X-ray emission, originating in a permeating coronal gas. This HIM is likely filling most of their ISM. This has consequences on the dust properties. This can be seen in Fig. 4.14.a, looking at a classic scaling relation between the dustiness and the specific gas mass 6, sMgas MgasM. Most ETGs appear to be distributed on a vertical branch, below the main trend. They appear to be depleted in dust, at a given specific gas mass. Investigating the contribution of the X-ray emitting coronal gas, we have displayed the specific dust mass, as a function of the X-ray-luminosity-to-dust-mass ratio, LXMdust, in Fig. 4.14.b (G21). The LXMdust ratio quantifies the X-ray photon rate per dust grain. We see that ellipticals occupy the lower right corner of this relation: they have a high photon rate per dust grain and a low specific dust mass. We have just shown that grains in a hot gas have a short lifetime (Eq. 4.30). The correlation of Fig. 4.14.b is thus likely the result of enhanced thermal sputtering in ETGs.

PIC PIC
Figure 4.14: Evidence of thermal sputtering in elliptical galaxies. Both panels show the results of the SED fitting of 800 nearby galaxies (G21), that we have started to discuss in Sect. 3.1.2.2. We have represented the measurements as Skewed Uncertainty Ellipses (SUE; Appendix F of  Galliano et al.2021). Each SUE corresponds to one whole galaxy. Galaxies are color coded according to their Hubble stage, T: (i) ETGs (T 0) in red; (ii) LTGs (0 < T < 9) in green; and (iii) Irregulars (T 9) in blue. Panel (a) shows how the dustiness scales with the specific gas mass. We see that most ETGs (red) are distributed along a vertical branch, below the main trend. Panel (b) shows how the specific dust mass varies with X-ray luminosity, LX, over dust mass, Mdust. ETGs occupy the lower right corner of this relation. Licensed under CC BY-SA 4.0.
4.2.2.3 Destruction by SN Blast Waves

We now focus on the effect of kinetic sputtering. This process leads to erosion and vaporization of grains in SN II blast waves. As we will see, this happens to be the major dust grain destruction mechanism. The kinetic sputtering rate is similar to the thermal case (Eq. 4.30), except that the sputtering yield now depends on the shock velocity, vs (cf. Fig. 4.13.b):

1 τsputkin(a,vs,nH) = 3nH a Y sput(vs). (4.31)

In addition, grain shattering in grain-grain collisions is an important dust destruction mechanism in SN II blast waves (e.g. Kirchschlager et al.2021).

Evidence in resolved regions. Although the efficiency of the process is debated, the reality of dust destruction by SN II shock waves is rather consensual. This process can even be observed in spatially-resolved SNRs. In particular, ISO and Spitzer MIR spectra of pre-shock and post-shock matter show systematic differences in, for instance: (i)  3C391 (Reach et al.2002); (ii)  SN1987A (Dwek et al.2008Arendt et al.2016); and (iii) PuppisA (Arendt et al.2010). The post-shock ISM exhibits:

Global model prescription. Similarly to what we did for grain growth (Eq. 4.23), it is convenient to express the dust destruction rate as a function of global galactic quantities. Such a formula was proposed by Dwek & Scalo (1980):

1 τSNdest(t) = mgasdest Mgas(t)RSN(t), (4.32)

where RSN(t) is the SN II rate (Eq. 4.11), and mgasdest is an empirical parameter quantifying the destruction efficiency. The latter represents the gas mass swept by a single SN II blast wave, within which all grains are destroyed. Eq. (4.32) can be understood the following way.

1.
A single SN II destroys a mass Zdust × mgasdest of dust.
2.
Knowing the SN II rate, the dust mass destroyed per unit time is therefore:
dMdust dt dest = Zdustmgasdest × R SN(t). (4.33)
3.
The dust destruction rate is then simply 1τSNdest = dMdustdtdestMdust, which gives Eq. (4.32).

The destruction efficiency. The dust destruction efficiency, quantified by the parameter mgasdest in Eq. (4.32), ranges in the literature between 100M and 1000M. It can be roughly estimated with the following arguments (Draine2009, slightly adapting his numbers):

1.
In a ISM of density nH 1 cm3, a SN II, with energy E0 1043 J (1051 erg), produces a blast wave that stays in the Sedov-Taylor phase (adiabatic expansion), until reaching a velocity vs 200 km/s (Eq. 39.22 of  Draine2011).
2.
Grains are primarily destroyed in the Sedov-Taylor phase. At the end of this phase, the radius of our blast wave is RSedov 24 pc (Eq. 39.21 of  Draine2011), corresponding to a total gas mass of MSedov 1900M, which is a rough estimate of our efficiency parameter, mgasdest.
3.
At solar metallicity (using the dustiness of Table 2.4), this corresponds to Mdustdest 10M .

This estimate corresponds to a dust lifetime of τSNdest 370 Myr, in the MW. This is a value close to what is found by more detailed, theoretical studies (e.g. Jones et al.1996). Note however that a recent re-estimate, using hydrodynamical simulations, and accounting for the role of dust mantles found: (i) a shorter lifetime for carbon grains; but (ii) a significantly longer lifetime (τSNdest 2 3 Gyr) for silicates (Slavin et al.2015). We will give our own take on this timescale in Sect. 4.3.1.2.

 A single SN II blast wave can destroy up to 10M of dust, at Solar metallicity, resulting in a dust lifetime of τSNdest 400 Myr.

4.3 Cosmic Dust Evolution

Cosmic dust evolution is the modeling of dust evolution from a global point of view, at the scale of a galaxy, over cosmic times 7. At galaxy-wide scales, most dust evolution processes can be linked to star formation: (i) formation of molecular clouds and their subsequent evaporation; (ii) stellar ejecta; (iii) SN shock waves; (iv)  UV and high-energy radiation. The characteristic timescale of these processes is relatively short (of the order of the lifetime of massive stars; τ 10 Myr) and their effect is usually localized around star-forming regions. For these reasons, the sSFR is an indicator of sustained dust processing. However, the dust lifecycle is a hysteresis. There is a longer term evolution, resulting from the progressive elemental enrichment of the ISM, which becomes evident on timescales of 1 Gyr. This evolutionary process can be traced by the metallicity. Fig. 4.15 illustrates these two timescales by comparing the SEDs of a few galaxies.

PIC

Figure 4.15: Effects of dust evolution on the SEDs of galaxies.. Each panel displays the observations and the SED model of two nearby galaxies (Rémy-Ruyer et al.2015), on top of the SED of the diffuse Galactic ISM (in grey). The red curve in panel (a) shows a quiescent Solar-metallicity galaxy. Apart from the stellar continuum, it is identical to the diffuse ISM. In contrast, the blue curve represents a low-metallicity quiescent system. Its dust properties are notably different: (i) weak or absent UIBs; (ii) overall hotter dust (FIR peaks at shorter wavelengths); (iii) a somehow broader FIR spectrum, resulting from a distribution of starlight intensities and/or an overabundance of small grains (cf. Sect. 3.1.2.2). This SED is qualitatively similar to the SED of a compact H II region (e.g. Peeters et al.2002b). The red curve in panel (b) shows a Solar-metallicity galaxy with a sustained star formation activity. Compared to its quiescent counterpart, it has a much hotter and broader FIR emission, originating at least partly in bright PDRs. The starbursting low-metallicity galaxy (blue curve in panel b) has the same features as its quiescent counterpart, with a broader FIR emission. Licensed under CC BY-SA 4.0.

4.3.1 Constraining the Dust Build-Up in Galaxies

We start by focussing on the build-up of the total dust mass. We will discuss the evolution of small carbon grains in Sect. 4.3.2. For terminological consistency of the rest of the discussion, let’s define the following metallicity regimes:

Very low metallicity:
Z 0.2 × Z;
Low metallicity:
0.2 × Z Z 0.45 × Z;
Normal metallicity:
Z 0.45 × Z.

We will see in Sect. 4.3.1.2 that these ranges correspond to dust evolution regimes of nearby galaxies.

4.3.1.1 Cosmic Dust Evolution Models

The first model accounting for the evolution of the gas content of galaxies and its cycle with star formation was presented by Schmidt (1959). The Eqs. 7-9 of Schmidt (1959) are the basic equations for the evolution of the gas mass as a result of the successive waves of star formation. Subsequent studies included the heavy element enrichment of the gas, therefore accounting for the chemical evolution of galaxies (e.g. Audouze & Tinsley1976, for an early review). Dwek & Scalo (1980) then initiated the first cosmic dust evolution model, by including grain processing in the gas enrichment modeling. Dwek (1998) modeled the radial trends in the MW, accounting for the individual elemental yields by stars of different initial masses. Such models have since then been refined (e.g. Morgan & Edmunds2003Dwek et al.2007Zhukovska et al.2008Galliano et al.2008aHirashita & Kuo2011Asano et al.2013Rowlands et al.2014Zhukovska2014Feldmann2015De Looze et al.2020Galliano et al.2021). These models are nowadays used to account for subgrid physics in numerical simulations of galaxy evolution (e.g. Hou et al.2017Aoyama et al.2020).

Physical ingredients and assumptions. The different cosmic dust evolution models we have just cited above all have differences. They however have a common set of physical ingredients and assumptions. In what follows, we describe the model of G21, which is well representative of the diversity found in the literature.

Star-formation regulation:
the evolution is driven by the SFH of the galaxy (cf. Sect. 4.1.1.3), considered as a box, where the mixing of freshly injected elements and grains is assumed instantaneous. This box can be closed or include the effects of infall and outflow. The infall and outflow rates are usually assumed to be proportional to the SFR:
Rin(t) δin × ψ(t) Rout(t) δout × ψ(t). (4.34)
Stellar evolution and ejecta:
in each time interval, [t0,t0 + Δt], a mass ψ(t0)Δt of stars is formed. The fraction of stars of different initial masses, m, is given by the particular IMF we have assumed (usually Salpeter or Chabrier; cf. Sect. 4.1.1.3). These stars have different lifetimes, τ(m) (cf. Fig. 3.18). They return to the ISM a fraction of their gas, freshly formed heavy elements and dust grain seeds, after a time t0 + τ(m). This is called the delayed injection process. After this time t0 + τ(m), a fraction of the initial stellar mass is locked in a remnant (white dwarf, neutron star or black hole; e.g.  Ferreras & Silk2000).
Dust evolution:
grain sources and sinks are estimated with the formulae we have discussed earlier in this chapter:
Stardust condensation
is accounted for by assuming a mean fraction of ejected dust by LIMS and SN II: (i) for LIMS, we have used δLIMS = 15% (Eq. 4.13); (ii) for SN II, we use Eq. (4.14).
Grain growth
is accounted for by Eq. (4.23).
SN blast wave destruction
is accounted for by Eq. (4.32).
Astration
is simply the fraction of dust consumed by SF, at a rate Zdust(t) × ψ(t).

The ejected masses. The dust evolution differential equations we will discuss below depend on the gas, heavy element and dust masses ejected by stars, at time t:

e(t) = min [m(tτ(m))]m+ m r(m) × B(t τ(m)) × ϕ(m)dm eZ(t) = min [m(tτ(m))]m+ Y Z(m) × B(t τ(m)) × ϕ(m)dm edust(t) = min [m(tτ(m))]mSN Y Z(m)δLIMS × B(t τ(m)) × ϕ(m)dm + min m(tτ(m)),mSNm+SN Y SN× B(t τ(m)) × ϕ(m)dm. (4.35)

These three equations are essentially the integral of the products of three terms: f(m) × B(t τ(m)) × ϕ(m)dm.

1.
The term f(m) is the mass of gas, heavy elements and dust ejected by a star of initial mass, m. For the gas mass, we have subtracted the remnant mass, r(m). The term Y Z(m) is the total heavy element yield (cf. Fig. 4.7).
2.
The term B(t τ(m)) is the stellar birth rate, τ(m) ago. It gives the number of stars of initial mass m, dying at time t.
3.
The third term is the IMF, giving the number of stars per mass bin. It is the number density we integrate over.
4.
In Eq. (4.35), the lower bound of the integrals, m(τ = t τ(m)), is the mass of stars having a lifetime t τ(m). We thus take the minimum mass of stars dying at time t (assuming star formation has started at t = 0). Low-mass stars (m 0.9M) are irrelevant for chemical evolution, as their lifetime is longer than the age of the Universe.

The equations of evolution. The physical ingredients and assumptions we have discussed earlier in this section translate into four coupled differential equations describing the temporal evolution of the stellar, gas, heavy element and dust masses, M, Mgas, MZ and Mdust:

dM dt =ψ(t) e(t) (4.36) dMgas dt = ψ(t) + e(t) + Rin(t) Rout(t) (4.37) dMZ dt = Z(t)ψ(t) + eZ(t) + 0 × Rin(t) Z(t)Rout(t) + Mdust τSNdest(t) Mdust τgrow(t) (4.38) dMdust dt = Zdust(t)ψ(t) astration + edust(t) ejecta + 0 × Rin(t) infall Zdust(t)Rout(t) outflow Mdust τSNdest(t) SNdestruction + Mdust τgrow(t) grain growth. (4.39)

Eqs. (4.36-4.39) simply express the time derivative of the mass, on the left-hand side, and the sum of the different individual rates on the right-hand side, some positive, some negative. We can note a few points.

1.
We have neglected metallicity variations in the total gas mass (Eq. 4.37), as it represents at most 2%.
2.
We have assumed that the infalling gas was free of heavy elements and dust. It is therefore 0 in Eqs. (4.38-4.39).
3.
On the contrary, the outflowing gas is carrying away heavy elements and dust. This is the reason why we loose these quantities at rates Z.Rout and Zdust.Rout, respectively.
4.
Dust destruction by SN II removes mass to the dust content, at a rate MdustτSNdest, but returns it as gas-phase heavy elements.
5.
This is the opposite for grain growth which removes mass to gas-phase heavy elements, at a rate Mgrowτgrow, and puts it into grains.

Dust evolution tracks. We now present a set of solutions to Eqs. (4.36-4.39). For simplicity, we adopt the Chabrier IMF (Eq. 4.6) and the delayed SFH (Eq. 4.9). The total baryonic, initial mass of the galaxy is assumed to be Mini = 4 × 1010M . Fig. 4.16 shows the time evolution of the main quantities, for a MW-like galaxy. We have adopted the following parameters:

τSFH = 3Gyr (Eq. 4.9) ψ0 = 20Myr (Eq. 4.9) δin = 0.05 (Eq. 4.34) δout = 0.05 (Eq. 4.34) Y SN = 0.007MSN (Eq. 4.14) 𝜖grow = 1000 (Eq. 4.23) mgasdest = 1000M SN (Eq. 4.32). (4.40)

Fig. 4.16.a shows the SFH we have adopted. The time evolution of the individual quantities are represented in Fig. 4.16.b. We can note the following points.

1.
The total baryonic content is conserved: Mini = M + Mgas. The stellar mass is increasing with time, while the gas gets depleted. In this particular simulation, they cross over around tcross 5 Gyr.
2.
The mass of heavy elements follows the stellar mass at t tcross, two orders of magnitude lower, as it is controlled by the accumulative stellar enrichment. Above t tcross, the net mass of heavy elements decreases due to astration, as the ISM becomes more tenuous.
3.
The dust mass follows the trend of heavy elements two orders of magnitude lower, for t 3 Gyr, as it is dominated by dust production by SN II. Around t 3 Gyr, the metallicity is high enough to render grain growth efficient. This is what Asano et al. (2013) have conceptualized as the critical metallicity. This is an important quantity, that depends on the SFH. It roughly delineates the two regimes dominated by SN II production and grain growth in the ISM. In the particular case of Fig. 4.16, it is Zcrit Z3. Above t tcross, dust production is dominated by grain growth, and MdustMZ 0.5, which is roughly the Galactic dust-to-metal mass ratio (cf. Table 2.4).
PIC
Figure 4.16: Dust evolution tracks for a MW-like galaxy. Panel (a) displays the adopted SFH. Panel (b) shows the evolution as a function of time of the four quantities of Eqs. (4.36-4.39). These quantities approximately reach the MW values at t = tpresent. The parameters are those of Eq. (4.40). Licensed under CC BY-SA 4.0.

Effects of the individual parameters. We end this section by demonstrating the effects of the seven parameters of Eq. (4.40) on dust evolution. We vary each parameter, one by one, keeping the other ones to their values in Eq. (4.40). We represent the dustiness, as a function of: (i) metallicity; (ii) sSFR; and (iii) gas fraction, fgas Mgas(M + Mgas). For completeness, we have represented the four SFH-related parameters in Fig. 4.17. Our center of interest is however the three dust evolution tuning parameters, represented in Fig. 4.18.

The SN II dust yield,
Y SN (Eq. 4.14), has essentially a scaling effect on the dustiness, below the critical metallicity, and has no effect above (Fig. 4.18.a-c). This is the reason why constraining this parameter requires very-low-metallicity objects. In the very-low-metallicity regime, dust and heavy elements directly come from stellar ejecta. Both quantities are thus roughly proportional, using our simple prescriptions. When Y SN 0.003M/SN, this parameter is so low, that stardust injection by LIMS becomes dominant, below the critical metallicity.
The grain growth efficiency,
𝜖grow (Eq. 4.23), has also a scaling effect on the dustiness, but above the critical metallicity (Fig. 4.18.d-f). It has no effect at very low metallicity, because 1τgrow Z (Eq. 4.23 with Z » Zdust). The tracks for the lowest values of 𝜖grow (in red on Fig. 4.18) exhibit a quasi-linear dustiness-trend with metallicity, over the whole range 8, because grain growth never reaches the efficiency of stardust condensation, in these extreme cases. The timescale ratio between SN II production and ISM grain growth is (using Eq. 2.2, Eq. 4.10, Eq. 4.11, Eq. 4.14 and Eq. 4.23):
τSNcond τgrow = 𝜖growm fSNY SNZdust(Z Zdust). (4.41)

Since we usually have Z 2 × Zdust, we can write the rough proportionality: τSNcondτgrow Z × Zdust, with Zdust being a steep function of Z. This is the reason why the relative efficiency of the two processes is so strongly metallicity dependent.

The SN II destruction efficiency,
mgasdest (Eq. 4.32), has the effect of decreasing the dustiness, at low to normal metallicity (Fig. 4.18.g-i). It however has no noticeable effect at very low metallicity. This is because the average dust mass destroyed by a single SN II blast wave is mdustdest = Z dust × mgasdest and Zdust « 1.
PIC
Figure 4.17: Effects of SFH-related parameters on dust evolution. Each row of panels represents the evolution of the dustiness as a function of metallicity, sSFR and gas fraction. In each row, we vary a particular parameter, whose values are given in the right panel. The other parameters are kept to their values in Eq. (4.40). Licensed under CC BY-SA 4.0.

PIC

Figure 4.18: Effects of tuning parameters on dust evolution. This figure is similar to Fig. 4.17, except that we have varied the dust evolution tuning parameters. Licensed under CC BY-SA 4.0.
4.3.1.2 Empirical Inference of Dust Evolution Timescales

We now discuss some empirical estimates of the three main grain evolution parameters (Y SN, 𝜖grow and mgasdest), derived from fitting dust scaling relations. This is not a new topic. Several studies have attempted to tackle this issue (e.g. Lisenfeld & Ferrara1998Morgan & Edmunds2003Galliano et al.2008aMattsson & Andersen2012Rémy-Ruyer et al.2014De Vis et al.2017Nanni et al.2020De Looze et al.2020). We have recently published such a study (G21). We will try to demonstrate the progress it has brought to the field.

Fitting dust scaling relations. We have used the SED modeling results of the DustPedia/DGS sample we have already discussed in Fig. 3.17 and Sect. 4.2.2.2 (G21). These results were obtained using the composite approach (Sect. 3.1.2.2), with the THEMIS grain properties, and a hierarchical Bayesian model (Sect. 5.3.3). These results are an estimate of: (i) the dust mass, Mdust, for all galaxies (G21); (ii) the stellar mass, M and SFR for most of them (Rémy-Ruyer et al.2015Nersesian et al.2019); (iii) the metallicity, Z, for about half the sample (Madden et al.2013De Vis et al.2019). The dust evolution model of Sect. 4.3.1.1 has then been fitted to our estimates of M, Mgas, Mdust, Z and SFR. We have adopted a hierarchical Bayesian approach (cf. Sect. 5.3.3), varying the following set of parameters.

The SFH-related parameters,
τSFH (Eq. 4.9), ψ0 (Eq. 4.9), δin (Eq. 4.34) and δout (Eq. 4.34) are varied individually for each galaxy. In other words, we have assumed that each galaxy has a particular, independent SFH.
The dust evolution tuning parameters,
Y SN (Eq. 4.14), 𝜖grow (Eq. 4.23) and mgasdest (Eq. 4.32) were varied, assuming they were common to every galaxy. In other words, we have assumed that the efficiencies of the dust evolution processes were universal.

Fig. 4.19 shows the fitted dustiness-metallicity relation, from our study (G21). We have represented the estimated observed quantities (SUEs) on top of the posterior PDF of our dust evolution model. This figure clearly shows the physical origin of the three regimes we had arbitrarily defined at the beginning of Sect. 4.3.1.

1.
At very low metallicity (Z 0.2 × Z), grain production is dominated by condensation in SN II ejecta. We see a roughly-linear dustiness trend with metallicity, with a rather low dust-to-metal mass ratio (ZdustZ 104):
Zdust Zdust 104 × Z ZforZ 0.2 × Z. (4.42)
2.
At low metallicity (0.2 × Z Z 0.45 × Z), we are in what we have called the critical metallicity regime. The dustiness rises sharply with metallicity, as grain growth in the ISM kicks in. The critical metallicity regime of individual galaxies is usually narrower (cf. Figs. 4.17 – 4.18). It is broadened here, because the PDF is the superimposition of all the galaxies, having different SFHs.
3.
At normal metallicity (Z 0.45 × Z), we have another roughly-linear dustiness trend with metallicity, sustained by grain growth in the ISM, with a Galactic dust-to-metal mass ratio (ZdustZ 0.5):
Zdust Zdust Z ZforZ 0.45 × Z. (4.43)
PIC
Figure 4.19: Dustiness-metallicity relation fitted with a dust evolution model. The SUEs are the results of the SED fitting of the DustPedia/DGS sample. Each SUE represents one galaxy, color-coded according to its type (cf. Fig. 4.14). We have overlaid the posterior probability distribution of the dust evolution model of Eqs. (4.36-4.39) as a yellow-orange density. This fit was used by G21 to infer the values of the three dust evolution tuning parameters. This is only a sub-sample of our 800 galaxies, as not all of them had reliable metallicity measurements. Licensed under CC BY-SA 4.0.

Evolutionary timescales as a function of metallicity. The dust evolution fitting of Fig. 4.19 allowed us to infer the values of the three tuning parameters. Accounting for possible systematic biases, we concluded the following (G21):

Y SN 0.03MSN,𝜖grow 3000,mgasdest 1200M SN. (4.44)

These efficiencies can be translated into timescales of the three dust evolution processes, in each galaxy. We have represented these timescales as a function of metallicity, in Fig. 4.20. The derived timescales for the MW are represented as a yellow star, although they were not used in the fit. We note the following points

1.
The timescale for dust condensation in SN II ejecta rises very abruptly with metallicity (cf. Fig. 4.20.a). It is realistic (i.e. shorter than the age of the Universe) only for very-low- and low-metallicity systems. At normal metallicity, another process needs to be invoked. Eq. (4.41) can be approximated by (assuming our limits in Eq. 4.44 are close to the true values):
τSNcond τgrow 1000 × Z Z × Zdust Zdust. (4.45)
At low very metallicity
(Z 0.2 × Z), Eq. (4.42) gives τSNcondτgrow 0.1 × (ZZ)2 « 1, which is another way to show that grain growth is inefficient in this regime.
At normal metallicity
(Z 0.45 × Z), Eq. (4.43) gives τSNcondτgrow 1000 × (ZZ)2 » 1, which is another way to show that grain growth is now dominant.
2.
The grain growth and blast wave destruction timescales (cf. Fig. 4.20.b-c) have rather similar features, because their ratio is (using Eq. 4.10, Eq. 4.11, Eq. 4.32 and Eq. 4.23):
τSNdest τgrow = 𝜖growm fSNmgasdest(Z Zdust) 4 × Z Z. (4.46)

The two processes thus balance each other around Zcrit Z4. This is where our value of the critical metallicity comes from. We note that, for the MW, we find τgrowMW 80 Myr and τSNdestMW 300 Myr, close to the values we had expected in Sect. 4.2.1.3 and Sect. 4.2.2.3. Yet, we did not put any prior on the Galactic values. This is a indication in favor of the consistency of our analysis.

3.
The ratio of the timescales for dust condensation in SN II ejecta and destruction by SN II blast waves is (using Eq. 4.14 and Eq. 4.32):
τSNcond τSNdest = mgasdest Y SN Zdust 500 × Zdust Zdust. (4.47)
At very low metallicity
(Z 0.2 × Z), Eq. (4.42) gives τSNcondτSNdest 0.5 × (ZZ) « 1, showing that SN II are net dust producers.
At normal metallicity
(Z 0.45 × Z), Eq. (4.43) gives τSNcondτSNdest 500 × (ZZ) » 1, showing that SN II are net dust destroyers.

 SN II are net dust destroyers, except at very low metallicity.

PIC
Figure 4.20: Empirical estimates of dust evolution timescales as a function of metallicity. The three panels represent the dust evolution timescales inferred from the fit of Fig. 4.19 (G21): (a)  τSNcond (Eq. 4.14); (b) τgrow (Eq. 4.23); and (c) τSNdest (Eq. 4.32). Each SUE corresponds to one galaxy. Licensed under CC BY-SA 4.0.

Methodological remarks. The study this section relies on (G21) was the first rigorous empirical determination of the dust evolution tuning parameters in Eq. (4.44), using a wide enough metallicity range to unambiguously constrain these quantities. We emphasize two important points.

Fitting dust evolution models is instrumental.
Numerous studies, trying to tackle the issues of cosmic dust evolution, simply overlaid dust evolution tracks (such as those in Figs. 4.17 – 4.18), on top of dust scaling relations (such as those in Fig. 4.14.a and Fig. 4.19). The issue with this approach is that two quantities of a given galaxy could be fitted with two different SFHs, at different ages. This is obviously inconsistent, but can give the appearance a good agreement with the data. We have avoided this pitfall with our hierarchical Bayesian approach. It allows us to avoid mutually inconsistent explanations of different trends and correlations. Overall, performing a rigorous fit does not help getting better solutions, but it definitely helps avoiding bad ones.
Low-metallicity systems are crucial.
As we have shown in Fig. 4.18, the effect of dust condensation in SN II ejecta can only be probed at very low metallicity. It is therefore necessary to have a good enough sampling of this regime. Without a good coverage at very low metallicity, the solution will consequently be degenerate. It will be impossible to disentangle the contributions of grain growth and stardust production. The relevance of dwarf galaxies here is not necessarily that they can be considered as analogs of primordial distant galaxies, but that they sample a particular, key dust production regime.

The controversy about stardust. We have opened this chapter with a citation of D09, about the belief that ISD could be mainly stardust. It is unfortunate that this discussion can sometimes turn into an ideological debate, in the literature.

In the nearby Universe,
for instance, De Looze et al. (2020) recently tried to show that SN II could be net dust producers at normal metallicity. They used a rather similar approach to ours. The only difference is that they did not have very-low-metallicity constraints. Their results were therefore clearly degenerate, but they forced their interpretation in favor of stardust. We will come back to these issues, from an epistemological point of view, in Chap. 5.
In the distant Universe,
this debate is also vigorous, as dusty galaxies are found at high redshifts (z 6; e.g.  Dwek et al.2007Valiante et al.2009Dwek et al.2014Laporte et al.2017). At this time, we are only 400 Myr after the reionization. We thus need a rapid source of dust, and stardust is seriously considered. However, our results confirm that grain growth can happen on timescales shorter than 100 Myr, provided that the ISM has been enriched by a first generation of stars, up to the critical metallicity. This therefore provides a simple solution to this conundrum.
In between,
measurements in absorption by Damped Lyman-Alpha systems (DLA), along the sightline towards a QSO or a Gamma-Ray Burst (GRB), can be used to estimate the metallicity and depletion of some elements in these systems, by comparing volatile and refractory abundances (e.g. De Cia et al.2016). These data produce a quasi-linear dustiness-metallicity trend, much flatter than Fig. 4.19 (e.g. Fig. 10 of Galliano et al.2021). If this trend is correct, it is consistent with stardust production at all metallicities. It is however difficult to understand the discrepancy between these systems and nearby galaxies. G21 conjectured it was possible that these estimates could be biased due to the dilution of heavy element absorption lines in near-pristine clouds, along the sightline, within the same velocity bin.

To try to rise above a mere ideological debate, we should not lose sight of the big picture, as the truth is the whole. Table 4.2 summarizes the observational evidence in favor of one scenario and the other.




Stardust ISM

origin origin



Elemental depletions (cf. Sect. 2.2.3 Sect. 4.2.1.1)




Nearby galaxy dustiness-metallicity trend (cf. Sect. 4.3.1.2)




Individual SNRs (before the reverse shock; cf. Sect. 4.1.2.2)

( )



Individual SNRs (accounting for the reverse shock; cf. Sect. 4.1.2.2)




Isotopic ratios of IDPs in meteorites (cf. Sect. 4.1.2.2)




Distant dusty galaxies (cf. Sect. 4.3.1.2)

( ) ()



DLA dustiness-metallicity trend (cf. Sect. 4.3.1.2)

( )



ISD is mainly amorphous, while CSD is crystalline (cf. Sect. 4.1.2.2)

()



Emissivity variation as a function of ISM density (cf. Sect. 4.2.1.1)

()



Table 4.2: Summary of the observational evidence about interstellar dust origin at normal metallicity. Check marks between parenthesis indicate uncertain evidence.

Limitations of our approach. Although our approach was successful in providing a unique, rigorous estimate of the dust evolution tuning parameters, and in deriving timescales as a function metallicity, it has several limitations.

Systematic uncertainties of the data
could bias the observed dustiness of our galaxy, displayed in Fig. 4.19. We have estimated the different potential biases on both the dust and gas mass estimates (Sect. 4.1.3 of G21). At normal metallicity, our measurements are consistent with the MW. At very low metallicity, we could suffer from: (i) the possible overestimate of the gas mass because of the extended gas halo of dwarf galaxies (Sect. 3.1.3.1); (ii) the systematic variation of the grain opacity with metallicity (Sect. 4.2.1.1); (iii) the potential overabundance of small grains at very low metallicity (Sect. 3.1.3.4); and (iv) the possible presence of unaccounted for VCD (Sect. 3.2.2.1). G21 concluded that these biases can not, in total, be larger than a factor of 4, which is consequent, but not sufficient to produce a linear dustiness-metallicity relation, that would be consistent with SN II dust production at all epochs.
The universality of the tuning parameters
is a questionable assumption. As we have shown in Sect. 4.2, all these parameters hide information about, in particular: (i) the typical grain size distribution; (ii) sticking coefficients, which are dependent on grain structure and composition; (iii) the topology of the ISM; (iv) stellar evolution. Exploring these variations with environment is however premature. We need independent estimates of the different factors we have just listed.
The simplicity of the dust evolution model
can cause discrepancies when trying to account for the observed SFRs. In particular, our model failed at reproducing the trend of sMdust with sSFR (cf. Sect. 5.2.3 of Galliano et al.2021). The likely explanation is that our parametric SFH is too simple. This could be solved by adding another SF component, to account for a potential recent burst.

4.3.2 Evolution of the Aromatic Feature Carriers

We close this chapter with a discussion about the trend followed by the grains carrying the aromatic feature emission. We have already discussed this point in Sect. 3.2.1.4, from a spectroscopic point of view. We now give a more general point of view, based on SED modeling, and discuss the different scenarios. We remind the reader that aromatic features can be emitted by PAHs 9 or small a-C(:H)s. This is a debated modeling choice (cf. Sect. 2.3). In the present section, we will assume that small a-C(:H)s are the carriers. Their mass fraction is qAF (cf. Sect. 2.3.3.1). Small a-C(:H) and PAHs emit similar aromatic feature strengths if qPAH 0.45 × qAF (G21).

4.3.2.1 The Different Evolution Scenarios

Aromatic features are significantly weaker in low-metallicity systems, compared to normal galaxies (cf. Sect. 3.2.1.4). This fact could indicate an increasing formation efficiency as a function of Z. However, low-metallicity environments have also their ISM bathed with a hard, permeating ISRF (cf. Sect. 3.3.2.3). Knowing that small a-C(:H)s are massively destroyed by such an ISRF (cf. Sect. 4.2.2.1), this trend could result from the increased suppression of aromatic features at low metallicity. Several scenarios have been proposed to explain these trends.

Enhanced destruction at low metallicity. Madden et al. (2006) proposed that small a-C(:H)s are more efficiently destroyed at low metallicity. This was supported by the relation in Fig. 3.41, between the strength of the aromatic features and the [Ne III]15.56μm/[Ne II]12.81μm tracing the hardness of the ISRF. The fact that dwarf galaxies have in general harder, more intense ISRF is linked to the following facts.

Another mechanism, proposed by O’Halloran et al. (2006), is the destruction of aromatic feature carriers by SN II blast waves. However, this scenario is less satisfactory, as blast waves tend to destroy all dust species (e.g. Reach et al.2002). They therefore do not constitute a consistent explanation for the selective destruction of small a-C(:H)s.

Inhibited formation efficiency at low metallicity. Several scenarios based on metallicity-dependent production mechanisms, in stellar ejecta or in the ISM, have been proposed.

The delayed injection of C,
by LIMS in their post-AGB phase, was proposed by Dwek (2005). Galliano et al. (2008a) conducted a quantitative comparison, using a dust evolution model comparable to Eqs. (4.36-4.39). We showed that it provided a consistent account of the observed trend of qAF with metallicity. We also took into account the small a-C(:H) photodestruction in H II regions, in the SED modeling, and estimated it was not sufficient to produce the trend. One major problem of this scenario is however that the volatility of small a-C(:H)s, that we see spatially, requires a mechanism to reform them in the ISM.
Shattering of large C grains
leads to the formation of small a-C(:H)s. Seok et al. (2014) have implemented this process in a cosmic dust evolution model and showed it could reproduce the qAF-Z trend. The range of SFHs required to cover the whole qAF Z trend is however wider than that needed to reproduce the dustiness-metallicity trend of the same sample (cf. our discussion in Sect. 4.3.1.2 about the importance of fitting dust evolution models).
Formation in molecular clouds
is another interesting scenario, as the molecular mass fraction is known to rise with metallicity (e.g. Schruba et al.2012). Greenberg et al. (2000) proposed that aromatic feature carriers could form on grain surfaces in molecular clouds and be photoprocessed in the diffuse ISM. Sandstrom et al. (2010) and Chastenet et al. (2019) showed that the spatial distribution of qAF is consistent with this scenario in the Magellanic clouds: qAF is higher in molecular clouds.

4.3.2.2 The Observed Trends

G21 have derived qAF in each galaxy of the DustPedia/DGS sample, that we have already amply discussed earlier in this chapter. Fig. 4.21 shows the evolution of this quantity with: (a) the metallicity, Z; and (b) the mean starlight intensity, U (Eq. 3.38).

PIC

Figure 4.21: Evolution of the mass fraction of small a-C(:H) grains with metallicity and starlight intensity. In both panels, we show the mass fraction of aromatic-feature-emitting grains, qAF (cf. Sect. 2.3.3.1) derived from the SED fit of G21. This quantity is displayed as a function of metallicity and mean starlight intensity, U (Eq. 3.38). Each SUE corresponds to one galaxy, color-coded according to its type (cf. Fig. 4.14). Licensed under CC BY-SA 4.0.

A better correlation with metallicity. Fig. 4.21.a shows a clear linear rising trend of qAF with metallicity (Eq. 9 of  Galliano et al.2021), and a decreasing trend of qAF, with U quantifying the intensity of the ISRF. Both correlations could be explained by any of the scenarios discussed in Sect. 4.3.2.1. We have however found that the correlation with metallicity is significantly better (cf. detailed discussion in Sect. 4.2.2 of Galliano et al.2021). This result is worth noting, especially since several studies focussing on a narrower metallicity range concluded the opposite (e.g. Gordon et al.2008Wu et al.2011). It probably relies on the fact the metallicities we have adopted in this study (De Vis et al.2019) correspond to well-sampled galaxy averages, while in the past a single metallicity, often central, was available and may have not been representative of the entire galaxy. This result suggests that photodestruction, although real at the scale of star-forming regions, might not be the dominant mechanism at galaxy-wide scales and that one needs to invoke one of the inhibited formation processes discussed in Sect. 4.3.2.1.

 At global scales, the mass fraction of small a-C(:H) seems to correlate better with metallicity than with the ISRF.

The global point of view. Overall, the qAF Z trend might have several origins. We think we can be confident about the following facts.

Small a-C(:H) photodestruction is real
and it is enhanced at low metallicity. It is however difficult to firmly establish if it is sufficient or not to explain the qAF Z trend.
The C/O ratio varies as a function of metallicity
(e.g. Pagel2003). For instance, in the SMC, (C/O) 14 ×(C/O). It means that if the a-C(:H)-to-metal mass ratio is Galactic, the abundance of small a-C(:H)s will be at most 1/4 Galactic.
In terms of filling factors of a multiphase ISM,
we can assume that small a-C(:H)s are absent of the HIM, WIM and H II regions, and are present in the other phases. Yet, the ISM of low-metallicity systems appears to be permeated with ionized gas, and their molecular cloud filling factor is lower.

What makes this question difficult to tackle is the diversity of spatial scales needed to properly balance the different processes. Ideally, we would indeed need to account for the following.

To know the origin of the qAF Z trend, we would need to reliably estimate the contribution to the integrated emission of both of these components, resolving sub-pc scales, in order to account for the enhanced emissivity in PDRs. This is out of reach of current facilities.

PIC
Figure 4.22: The potential of quiescent very-low-metallicity galaxies to understand the origin of small a-C(:H) grains. The data in both panels are identical to Fig. 4.21. We have simply added the hypothetical observations of a quiescent very-low-metallicity galaxy (solutions 1 and 2). The cyan stripe in panel (a) is the analytical fit of the trend given in Eq. (10) of G21. Licensed under CC BY-SA 4.0.

The potential of quiescent very-low-metallicity galaxies. An alternative would be to observe quiescent very-low-metallicity systems. We know such a population of galaxies exist (e.g. Lara-López et al.2013). Let’s assume that we can estimate qAF for a galaxy with Z 0.03Z and U 0.3. We have represented the two possible solutions on Fig. 4.22. We see that:

1.
if we find qAF 0.006, it will be consistent with the Z trend (Fig. 4.22.a), but not with the U trend (Fig. 4.22.b); or
2.
if we find qAF 0.17, it will be consistent with the U trend (Fig. 4.22.b), but not with the Z trend (Fig. 4.22.a).

Such observations would require a sensitive MIR-to-FIR observatory, such as what SPICA (van der Tak et al.2018) could have been.

1.In thermonuclear reactions, the high temperature gives nuclei enough kinetic energy to overcome their Coulomb barrier, and allows them to fuse with each other.

2.The Sun’s core is at T 1.5 × 107 K.

3.The degeneracy pressure is due to the fact that fermions can not occupy the same state. In very dense environments, this leads to a pressure: electron degeneracy pressure in white dwarfs; neutron degeneracy pressure in neutron stars.

4.Careful though, some authors quote IMFs, normalized as mϕ(m)dm = 1.

5.As a reminder, the metallicities of the Magellanic clouds are: ZLMC Z2 and ZSMC Z5 (Pagel2003).

6.Given an extensive quantity, Q, it is common, in extragalactic astronomy, to define the corresponding intensive specific quantity, sQ QM, by dividing by the stellar mass.

7.We assume that the age of the Universe is tpresent 14 Gyr (e.g. Hogg1999).

8.The sawtooth features in Fig. 4.18.d-f, for the two lowest values of 𝜖grow (red and orange), are numerical artefacts due to the fact that grain growth is so low, that SN II can clear dust faster than our time resolution.

9.Alternative acronym: Poor-people’s Amorphous Hydrocarbon...