In situ observation of the impact of hydrogen bubbles in Al–Cu melt on directional dendritic solidification

Much research has already been focused on the solid-bubble interaction in the interdendritic space for solidifying materials. However, commonly, bubble nucleation is not limited to the mushy zone but also occurs in the liquid melt. In the present research on an Al-10%wt.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10 \, \%\mathrm {wt. \,}$$\end{document}Cu alloy, the interaction between these bubbles and the approaching solidification front becomes apparent under in situ X-radiography and allows for new insights into the influence of bubbles on the solidifying microstructure. The observed effects comprise bulging of the solidification front toward the bubble, bending of dendrites in front of the bubble, coronal outgrowths surrounding the bubbles, as well as bubble growth, bubble pushing, and bubble eruption. It is found that for the present Al–Cu alloy, the local variation in the solidification speed can be attributed to the bubbles’ insulating properties. The range of this effect was observed to be up to 900μm\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$900 \,\upmu \text {m}$$\end{document}, depending on the bubble diameter, locally increasing solidification speed by up to 350%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$350 \, \%$$\end{document}. The influences of Marangoni vortices and coronal nucleation of misoriented dendrites around bubbles on the homogeneity of the microstructure are discussed. A comparison with experiments on model alloys and simulations from various other studies highlights the similarities and differences to this metallic alloy system.

Simulations, such as the work by Han et al. [2], predict wave-like nucleation and distribution of pores in the solidifying material or were developed to model the growth of various different types of porosity (microporosity, macroporosity, and shrinkage) simultaneously as shown in the research of Pequet et al. [3]. Even the development of large interdendritic pore networks and their relationship to the pore curvature were the focus of modeling by Felberbaum and Rappaz [11].
But not only simulations have yielded results on the topic of porosity-solid interaction; over the years, especially in situ imaging setups have established themselves as the primary means for observing the solidification process. Lee and Hunt, pioneering in situ X-radiography imaging for analysis of metallic samples in 1997, observed the formation of pores in aluminum (Al)-copper (Cu) alloys and characterized interdendritic pore morphology as a function of temperature [12]. A change of pore morphology was identified by Arnberg and Mathiesen and related to constraints by surrounding dendrites [4]. In 2016, Murphy et al. described the fragmentation of dendrites and ensuing columnar to equiaxed transition caused by bubbles in the mushy zone, as well as the effect of bubbles on intergranular spacing [13].
However, bubble nucleation is not limited to the interdendritic region. In the melt, impurities such as oxide particles can lead to heterogeneous nucleation of gas bubbles [14]. These bubbles can escape due to buoyancy, be engulfed by the dendritic front, or push away/re-melt the dendritic network [6,13,15]. Thus, the need arises to better understand the interaction between those pre-existing bubbles in the melt and the solidifying material. In particular, the advantages of in situ imaging compared to postmortem analysis, when examining such interactions at an early solidification stage, are apparent: The whole process can immediately be observed via the live-feed of the camera, clearly depicting the complete sequence of solidification events in real time. Simulations on metallic alloys and observation of transparent, organic model alloys have shown a notable influence of gas-bubbles on the solidification behavior of the material for this specific scenario of bubbles in front of the mushy zone.
Considering bubbles in the liquid in front of the solidifying material, Sekhar and Trivedi suggested localized variations in the solidification velocity due to concentration differences in the melt in the vicinity of pre-existing bubbles for a transparent organic model alloy [16]. A simulation of a metallic alloy by Catalina et al., on the other hand, predicted the dominant effect on the solidification speed to be thermal insulation: The reduced heat flow through the gas, when compared to the liquid metal, would cause the melt on the solid-facing side of the bubble to be colder than the melt without bubble, and thus induce a localized increase in solidification speed [17].
Other simulation studies advocate the development of Marangoni flow fields surrounding the bubble. This interfacial effect, driven by a gradient in surface tension around the bubble, transports hot melt toward the dendrite front, causing bending of the dendrites. Corresponding experiments on a transparent organic succinonitrile-based alloy were performed by Nabavizadeh et al. in microgravity aboard the International Space Station, so as to eliminate gravity-driven factors such as convective flow and buoyancy of bubbles. The study concludes that Marangoni flow indeed affects the growth direction of dendrites, causing them to bend away from the bubble [18].
A third type of influence on the microstructure was observed by Han and Trivedi in 1994 and discussed in-depth by Xing et al. in 2012, both for succinonitrilebased transparent alloys [19,20]. The solidifying material, upon coming into contact with the bubble, produced corona-shaped outgrowths along the gasmelt interface, which eventually evolved into dendrites growing along the temperature gradient.
With the focus on these three effects-bulging of the solidification front, deviation of dendrites due to Marangoni flow, and coronal outgrowths-the present work analyzes how pre-existing bubbles cause non-homogeneity of the growth and microstructure in an Al-Cu alloy. In situ X-radiography allows for direct observation of the directional columnar dendrite growth of the Al-10 %wt: Cu alloy sample processed in a Bridgman-Stockbarger-type furnace. To minimize buoyancy of the bubbles and grains, a sheet-like sample was processed in horizontal position, i.e., the gravity vector is orthogonal to the plane of the thin (near-2D) sample.
Different hydrogen (H) contents of the metallic samples have been achieved through the use of a gasloading furnace [21]. Deformation of the dendrite front, growth direction of the dendrites, and morphological homogeneity will be discussed, as will be correlations between bubble diameter, distance range of the insulating effect of bubbles and increase in solidification velocity near bubbles.
Similarities and differences between the present research and the aforementioned models will be addressed.

Sample preparation
The good contrast in X-radiography between the Aldendrites and the melt [15,22,23], as well as the alloy's property of H being the only soluble gas in it [14,24,25], ruling out chemical effects such as oxidation at the bubble-melt interface, are the underlying considerations for using the Al-Cu system.
More specifically, this study is based on an Al-10 %wt: Cu alloy, the element ratio at which the alloy is density-matched [26]. At this elemental composition, the solidified Al has nearly the same density as the surrounding melt, and thus the influence of buoyancy effects is minimized.
To achieve reproducible gas-bubble numbers, Al-10 %wt: Cu bulk samples were cast in a gas-loading furnace following the procedures described in [21]. Melting the alloy's raw materials in a well-defined environment (pressure, temperature) before introducing an atmosphere of controlled H-content, the liquid metal was then held for a defined amount of time at maximum temperature, before being cast in a steel-die acting as mold and heat-sink. By varying these four parameters (pressure, temperature, holding time at maximum temperature, and loading-gas concentration in the atmosphere), specimens with an H-content of 5:4 Á 10 17 atoms=g ð25:48 ppmaÞ and 2:7 Á 10 17 atoms=g ð12:70 ppmaÞ were produced. The gas contents for comparable Al-10 %wt: Cu alloys used in other studies vary between 5:3 Á 10 15 atoms=g and 5:4 Á 10 17 atoms=g, situating the present work toward the upper end of the H-content spectrum [11,12,21,27,28]. The divergence of values by up to two orders of magnitude can be attributed to different casting and cooling conditions or to diverse methods of evaluation and shows the necessity for a controlled gas-loading process, as was already discussed elsewhere [21].
A short overview over the state of the art of Hcharging of Al-alloys is given in Table 2 in the supplemental data, summarizing publications from 1989 to 2020 and situating the present work among them. Furthermore, as the literature sources give H-content in varying units [17,21,[29][30][31], Equations (1)(2)(3) in the supplemental data provide the conversion for results in different units (½atoms=g, ½ccm=100g STP , ½ppma, and ½ppmw).
To prepare the samples for the X-ray investigations, the cast bulk samples were cut at a distance of 10 mm from the bottom face of the casting mold and ground according to ASTM standard E3-1115 to dimensions of 5 mm x 50 mm x 150 lm [32]. The cutting was performed on a Buehler Isomet4000 Linear Precision Saw equipped with a Buehler MetAbrase abrasive cutoff wheel at 1700 rpm and a feed-rate of 1:2 mm Á min À1 . Subsequently, the sheet-like specimens were manually coated in boron-nitride (BN) spray (HeBoCoat 401E provided by Henze Boron Nitride Products AG) to avoid chemical interactions with the quartz-glass crucible. Due to the manual coating process, variations in the BN-layer thickness can cause darker patches overlaying the X-ray images. An in-depth description of the sample preparation process can be found elsewhere [21,33].

In situ directional solidification experiments
To allow for in situ observation of the microstructural solidification process, the experiments were carried out in a Bridgman-Stockbarger furnace featuring an X-radiography unit as shown in Fig. 1. Taking into account the detector's native pixel size of 9 lm, as well as the sevenfold magnification resulting from the distance ratio of source/detector to source/sample, the effective pixel size of the setup amounts to 1:3 lm. A more in-depth description of the experimental unit can be found elsewhere [33]. Due to the thin (150 lm) sample geometry, the microstructure can be distinctly observed, as the dendrites usually do not superimpose each other, while at the same time, convective flow and buoyancy are reduced. It has been shown that the geometrical constraints in the nearly two-dimensional specimen, if oriented horizontally, effectively restrict gravitational effects such as buoyancy on bubbles and dendrites, and, to a certain extent, convection [22,34].
It should be noted, however, that while horizontal orientation of the sample prevents bubble and dendrite movement caused by buoyancy effects and reduces thermal convection to a negligible level, according to Nguyen-Thi et al. it does not eliminate solutal convection to a point identical to real microgravity [22]. For stable thermal and solutal conditions, vertical upward solidification reduces solutal convection further but results in buoyancy effects acting on bubbles and dendrites. As the present study requires immobile bubbles, the horizontal orientation was chosen.
An air-atmosphere of 1013:25 hPa (1 atm) ensured sufficient pressure to slow the degassing of H down to resolvable levels while taking into account the chosen frame rate of 0:54 Hz. The constant temperature gradient across the 6-mm-wide adiabatic gap was established between two heaters (ceramic glow igniters encased in heat resistant steel) as described by Zimmermann et al. [33]. For the directional solidification experiments, the samples were first thermally stabilized for 10 min, followed by a horizontal displacement of the sample inside of the furnace (which is mounted in place) at defined velocities for each specimen as shown in Table 1.
To ensure statistical significance, a total of four distinct samples (two with a lower H-concentration and two with a higher H-concentration), processed at two different pulling speed and thermal gradient combinations, were analyzed over five melting/solidification cycles as can be seen in Table 1. Pulling speed and thermal gradient were not found to have significant influence on the behavior of bubbles and will not be further discussed in this work.

Results
During the solidification process, H-bubbles are observed to nucleate and remain in the melt in front of the solidifying material until they either pop or are engulfed by the solidification front. In the latter case, for about 25 % of these gas bubbles, deformation of the solid-liquid interface is observed in the immediate proximity of the bubble. This behavior was found in both, dendritic and planar fronts. Figure 2 shows a sequence of X-ray images taken from a specimen with an H-content of 2:7 Á 10 17 atoms=g, starting at the beginning of solidification. Development of the localized deformation continues until the bubble is half-engulfed by the crystallites. Growth of the gas bubble correlating with the decreasing distance to the solidification front can be observed: Bubbles far away from the solidification front also grow, but at a much slower rate than the bubbles close to the crystallizing solid, as seen in Fig. 3a and b. The gas-bubble is then pushed upwards, followed by sudden acceleration and eventually popping of the bubble. It has to be noted that the five described effects of the bubble growing, being pushed, being engulfed, accelerating, and popping are not commonly all occurring in an individual instance of the solidification front's bulging. In most observed cases, only one or two of the aforementioned phenomena would be observed.
As seen in Fig. 4a, relating the range of the bulging effect, quantified by d flat (the shortest distance between the bubble and the solid-liquid interface for which the solidification front is still flat and has not shown any deformation), to the bubble diameter D (the diameter at the moment of first deformation of the solidification front) in the aforementioned situation, it was found that with increasing bubble diameter, the range of the effect also increases. The increase in solidification speed (in %) of v max compared to the local mean solidification speed of the solidification front is plotted in Fig. 4b. The graph illustrates the relation to bubble size and bubble distance to the front: The larger the bubble and the closer to the growth front, the higher the growth speed. This effect can be observed for both H-concentrations of 5:4 Á 10 17 atoms=g and 2:7 Á 10 17 atoms=g. With higher H-content, the number of observable events increases accordingly.
Another phenomenon observed in this study, shown in Fig. 5, is the deviating/bending of dendrites prior to coming into contact with the H-bubble. Due to the high brightness of the bubble and the lower contrast of dendrites compared to the melt, this effect is most clearly observed on bubbles that pop while being engulfed by the solidification front.
However, instead of bending and bypassing the bubble, the dendrites can also be stopped by the bubble or come into direct contact with it at a certain distance off-center but without stopping. The bubble, as shown in Fig. 6a, upon experiencing immediate collision with a growing dendrite, then sees a coronashaped layer of outgrowths forming at the bubblemelt interface and growing into the melt. These outgrowths quickly begin to orient themselves and eventually develop dendrites either growing in the same direction as the initial crystallites or deviating and forming a new grain (see Fig. 6b). However, even for the case that no new grain develops, a circular region of about twice the bubble diameter surrounding the pore remains misoriented with respect to the columnar microstructure in its vicinity.
It should be noted that despite the original dendrites growing at an angle of about 45 , as seen in Fig. 6, the tips of all dendrites were observed to advance at the same speed, effectively resulting in an upward-growing solidification front.
To summarize the observations, the novelty of the presented findings in this work lies in the observation of three effects that had not previously been documented experimentally for metals: bulging of the solidification front close to bubbles in the melt, bending/deviation of dendrites around gas-bubbles, and coronal outgrowths nucleating around bubbles.
Additionally, five secondary phenomena appeared, notably bubble growth, bubble pushing, bubble engulfment, bubble ''eruption'', and bubble popping, that will be briefly explained in the following section, for the convenience of the reader.

Bulging of the solidification front
When discussing the deformation of a solidification front, the Mullins-Sekerka instability is often at the root of the effect: The solidification front randomly develops an irregularity/a bulge of atomic scale, which locally increases the thermal gradient and thus accelerates the growth of the initial bulge. This in turn leads to a self-amplifying scenario, which only slows down once the Gibbs-Thomson effect restricts the growth speed due to the surface curvature becoming too large to sustain for the system. That is what eventually results in the formation of dendrites [35,36]. However, in the present case, deformation of the solidification front does not appear randomly but is only observed close to bubbles. It shows no periodicity and affects a segment of the solidification  where the bubble is pushed away by the approaching solid, leaving behind a U-shaped dent in the bulged-out solidification front. Between t ¼ 72:2 s and t ¼ 74:1 s, the bubble suddenly accelerates and moves away from the surrounding crystalline material toward the hot end before popping. In all images, gravity acts orthogonally through the image plane.
front of up to several hundred microns for a single bulge, depending on the size of the bubble, as shown in Fig. 4. The impact of the bubble on the crystallizing solid can be explained as follows: Considering gas bubbles as particles with ideal insulating properties, the melt on the solid-facing side of the bubble is de facto colder than the melt at the same distance from the solid but far away from the bubble. This local temperature gradient will thus cause a deformation of the solidification front, in other words, an increase in growth speed of the crystallites closest to the bubble. The insulating effect and resulting temperature gradient both increase with bubble size. This is supported by the finding that larger bubbles have a greater impact on the growth speed increase as well as a bigger zone of influence (characterized by d flat ). This effect has already been discussed in multiple works focusing on simulation [17,37,38], and now, having observed it in situ, we can support these simulations with experimental evidence for a metallic material as shown in Fig. 2.
The discrepancy to Sekhar and Trivedi's results [16], who found that bubbles lead to a depression in the solid-liquid interface, can be explained by the difference in experimental setups: While both their work and the present study use glass crucibles (thermal conductivity in the range of 0:8 W Á ðm Á KÞ À1 [39]), their work focuses on succinonitrile-based organic alloys with a thermal conductivity of around 0:3 W Á ðm Á KÞ À1 [40], while the present work features an Al-Cu alloy (thermal conductivity around 80 W Á ðm Á KÞ À1 [41]). The heat flow through the sample itself is thus about 2.6 orders of magnitude higher in the present experiment than in Sekhar and Trivedi's study, and accordingly, the thermal effects through the sample are more pronounced. Another experimental difference is found in the alloy composition: Sekhar and Trivedi used a solution of 5 %wt: succinonitrile-95 %wt: water (succinonitrile-phase crystallizes), whereas the present alloy composition amounts to 90 %wt: Al-10 %wt: Cu (Al-phase develops). The amount of rejected solute upon solidification is thus more than nine times as large in Sekhar and Trivedi's experiments, and concentrational effects such as solute pile-up are more prominent in their study.
Except for the material used in the experiments, the geometry of the setup also influences the results to a certain point; specifically, the systematic error of 2D simulations and near-2D experiments needs to be kept in mind: as shown in Fig. 2b, for bubble diameters smaller than 150 lm (thickness of the sample), the increase in growth speed of dendrites compared (a) (b) Figure 3 The growth of bubbles from five different heating/solidification cycles was evaluated as a function of a the distance to the solid and b time. In all cases, the volume of Hbubbles increases with decreasing distance to the solidification front and with longer duration of the experiment. For the computation of the bubble volume from the 2D data, perfectly spherical bubble geometry was assumed for D\150 lm (thickness of the sample). For larger bubbles, the spherical caps were subtracted symmetrically. In sample 2, the observed bubble moved and subsequently shrank when the solidification front was about 200 lm away, before starting to grow again. Different amounts of solved H, specifically 5:4 Á 10 17 atoms=g (solid squares) and 2:7 Á 10 17 atoms=g (empty squares), do not affect this behavior, only the amount of observable events.
to the speed of the advancing solidification front becomes smaller with increasing bubble size. This effect can be attributed to the bubble acting as an obstruction for the Al-rich melt feeding the growing crystallites. For small bubbles, the melt can still flow past the bubble, on its sides as well as above or below it. The bigger the bubble, the smaller the gap between sample surface and bubble becomes. This, however, changes at D [ 150 lm, when the bubble connects the top and bottom faces of the sample. All melt is forced to flow around the sides of the bubble, and regardless of the bubble size, the flow rate stays the same from then on.
The bulging effect has been observed for 25 % of the bubbles in the present setup. This observation rate is attributed to the growth and popping events that happen throughout the experiment, as well as the geometrical position in thickness-direction of the bubbles relative to the growing dendrites that cannot be determined in this near-2D setup: If a bubble is adhering to the top side of the sample, but the dendrites grow closer to the bottom side of the sample, the growth direction is not blocked and only minimal interaction is possible.

Bending of dendrites
The observed bending of the dendrites' growth direction away from the bubble could be attributed to the Marangoni effect. The Marangoni effect is an interfacial phenomenon that induces fluid transport along an interface following a gradient in surface tension. The fluid is transported from a region of lower surface tension toward a region of higher surface tension. In the case of a spherical interface in an almost-2D setup, this would lead to the formation of two vortices on the outside of the sphere, as depicted in Fig. 7. The direction of these flow vortices depends on two main factors: The concentration gradient of the solute being different due to dendrite growth locally enriching the melt in one element (here Cu), and the temperature difference already mentioned in ''Bulging of the solidification front'' section.
While the temperature gradient will always cause a flow from the hot side of the bubble to the cold side, the concentration gradient-driven flow depends on the melt compositions and the corresponding surface tension distribution around the bubble. Depending on the thermal/concentrational gradients of the system, the flow vortices can thus transport fluid in either direction for different experimental setups [4,18,42,43].
In the case of Al-Cu alloys, the surface tension increases with both increasing Cu-content and decreasing temperature [44]. In the free melt (ahead of the mushy zone), the concentration of Cu is highest close to the solidification front. Additionally, on the dendrite-facing side of the bubble, the temperature is lower than on the melt-facing side, similarly causing the surface tension to be higher. As both parameters cause an increase in surface tension on the dendritefacing side of the bubble, Marangoni vortices will transport Al-rich, hot melt toward the dendrite front. This causes a deviation of the dendrites' growth direction away from the bubble under microgravity conditions, according to Nabavizadeh et al. [18].
Under standard gravity, convective effects as well as movement of the bubbles due to buoyancy make it challenging to observe the Marangoni effect on bubble interfaces [18]. As the near-2D setup of the experiment in horizontal position reduces both, convection and buoyancy-effects, Marangoni flow fields may be strong enough to influence the growing microstructure in an observable way. However, it cannot be determined with absolute certainty if the bending of the dendrites close to the bubbles, as observed in the present study, is due to the Marangoni flow or due to geometrical constraints. Moreover, a consistent picture regarding the impact of directional flow fields on dendrites has not yet emerged: Nguyen-Thi et al. found bending to occur in the direction of the flow, away from the feedingelement-rich melt, and toward the colder region [22]. Similar results were obtained by Nabavizadeh et al. in simulations [18]. On the other hand, Murakami et al. observed the bending of dendrites against the flow direction, toward the hotter region and toward higher feeding-element concentration [45].
A notable difference between those works, however, is the solidification speed of the alloy: While in Murakami's experiment the dendrites grew with around 1 mm Á s À1 , the crystallization speed in Nguyen's study was 1000 times slower [22,45]. As the solidification speed in the present case is of the same order of magnitude, the experiment of Nguyen et al. would well approximate our results, which is supported by the observation that the dendrites in the present research also bend in direction of the fluid flow (see Fig. 5).
It should be noted that the Marangoni effect and the bulging are not mutually contradictory, as the increase in growth speed due to the insulating effect of the bubble acts on a larger range than the Marangoni effect. The latter would thus only cause a slight depression at the tip of the bulge, coinciding with the zone of solute enrichment between bubble and dendrites, in which the high Cu-concentration retards the growth of Al-dendrites [17], the differentiating factor between the depression due to solute enrichment and the depression caused by Marangoni flow being the bending of the dendrites.  Figure 4 Three-dimensional projection color plot of a the maximum solidification front growth velocity and b the relative increase in solidification front growth velocity varying with d flat (xaxis) and D (y-axis). d flat is the distance between a bubble and the solidification front before it bulges out and D is the bubble diameter at the moment of first deformation of the solidification front. In both panels, different amounts of solved H, respectively, of 5:4 Á 10 17 atoms=g (solid squares) and 2:7 Á 10 17 atoms=g (empty squares), do not affect this behavior, only the amount of observable events. The color map consists of data points from all four samples. Different solidification parameters did not show any impact on the bubbles; thus, for the convenience of the reader, the authors refrained from using color coding on the data points.

Coronal growth
The engulfment of a gas bubble by a dendritic layer around the gas-liquid interface, similar to heterogeneous nucleation at a solid-liquid interface, was observed and discussed by Xing et al. in situ for organic succinonitrile-based transparent alloys [20]. The dendrite that comes into contact with the bubble is stopped, before a halo of nucleating crystallites engulfs the bubble. Eventually, the influence of the applied thermal gradient causes some of the extrusions to evolve into columnar dendrites that either grow in the direction of the temperature field while the remainder is blocked by the surrounding dendrites, or deviate from the direction of the original dendrites and instead grow a new grain starting from the pore. Our results confirm those findings to be valid not only for organic models, but even for our Al-Cu alloy, determining that the metallic nature of the experimental system does not invalidate this effect.
While pores are generally known as stress concentrators and starting points for cracks, literature tends to evaluate them based only on the size and shape of the pore itself [46]. In the present case, however, it can be seen that the influence of the pore extends on a microscopic scale to a halo-shaped area surrounding it, up to twice the bubble radius. In this region of the sample, the directional dendritic growth is replaced by a zone of disordered crystallization. Macroscopically, an alteration of the original grain structure can occur. This can compromise the mechanical stability if a homogeneous columnar microstructure is desired in an application.

Secondary reactions of the bubbles to the approaching solidification front
As the distance to the solid-liquid interface decreases, one can observe a range of secondary effects on the gas bubbles.
Similar to the phenomenon of interdendritic bubble growth during solidification, which has already been extensively discussed in other publications [6,7,11,12,37,47], the gas bubbles in front of the growing dendrites also grow quickly (see Figs. 2 and  3). The effects are the same here as in the aforementioned interdendritic case: Upon crystallization, the H-solubility in the material is greatly reduced, and the H gets rejected from the Al into the melt, where it diffuses toward points of locally lowered partial H-pressure. Existing bubbles thus grow by absorbing the surrounding H. In the interdendritic space, this   17 HÀatoms=g is observed. a The discontinuity in upward columnar dendrite growth, covering a ring-shaped area of twice the bubble diameter surrounding the gas-pore, forms following the direct contact of a dendrite with a H-bubble. Eventually, the outgrowths evolve into dendrites and grow larger. b The new dendrites can either grow in the same direction as the initial crystallites or deviate and form a new grain. In all images, gravity acts orthogonally through the image plane. effect is more pronounced than in the free melt, as the distance to the solid Al-dendrites (where the H rejection takes place) is on the scale of interdendritic spacing, while in the free melt, the distance between bubble and dendrites only decreases gradually upon advancement of the solidification front [37]. Moreover, when comparing bubbles in the melt, close to the solid, to bubbles farther away, multiple factors cause the nearby bubbles to grow faster: Close to the solidification front, the Cu-concentration is higher and, as explained by Anyalebechi [48], increased Cucontent decreases H-solubility, as do the lower temperatures at that position of the sample [49]. Eventually, H being rejected by the solidifying metal leads to a locally increased H-concentration in the melt. The sum of these effects, as well as the out-gassing over time of the supersaturated H, cause more H to agglomerate in the form of bubbles in the vicinity of the solidification front than far away from it. The distance to the solidification front is thus the indirect parameter that affects the growth rate through the direct parameters which scale with the distance. In both Fig. 3a and b, the effect of these combined factors can be observed: The growth rate of the bubble is not constant, as it would be for a purely time-dependent-but not spatially varying-process and instead increases the closer the solid gets to the bubble.
As the solidification front reaches the bubble, an effect analogous to instances of pushing and engulfment of solid particles can be observed, as gas-bubbles are also pushed or engulfed by the solidifying material (see Fig. 2) [50][51][52][53]. However, as the bubble itself is a fluid of very low density, the lift forces that are known to be responsible for particle pushing only apply marginally. These lift forces are the Saffman force and the Magnus force [50,[53][54][55][56]. The Saffman force acts on a particle in a flow field when there is a gradient in relative flow speed. This is the case when a particle in a flow field is close to a non-moving solid surface, as the flow speed decreases with decreasing distance to the surface. The particle is then pushed away from the surface by the Saffman force. The second lift force, the Magnus force, is generated when particles are rotating, which produces a lift force perpendicular to both the direction of the fluid flow and the axis of rotation. The rotation is induced by the forced convection in front of the dendrite tips due to gravity. The effects of convective flow are, however, strongly reduced for the case of a gas bubble as compared to that of a solid particle: Inducing rotation in a sphere of fluid inside of a second fluid requires transmission of shear forces, which are almost negligible in the case of a gas due to the gas' low viscosity and the reduced convective flow resulting from the horizontal orientation of the sample [50,[53][54][55][56]. Marangoni flow (dashed lines) surrounding a gas-pore, and its effect on the growth direction of the dendrites under microgravity. The differences in temperature and composition of the Al-Cu melt cause a gradient in surface tension around the bubble. According to the Marangoni effect, liquid flow toward the side of highest surface tension ensues and two flow vortices at the sides of the bubble form. This flow of hot, Al-rich melt causes the dendrites in proximity of the bubble to bend/deviate from their original growth direction. As adapted from [18,42].
The bubbles also sporadically display bursts of fast movement in a certain direction upon interacting with, and being partially engulfed by, the advancing solidification front (see Fig. 2 for t ¼ 74:1 s). As Han discovered for transparent cyclohexane model alloys, upon being at least partially surrounded by solidifying material, growing bubbles can experience deformation, resulting in an increase in internal pressure [57]. Once the internal pressure becomes large enough, the bubble will then push aside the crystallized material and escape the mushy zone while restoring its own energetically favored round shape. Han coined this process ''bubble eruption''.
It has to be noted, however, that Han's case and the present work are different in that Han's bubbles are confined to the mushy zone, while in the present experiments, the solidifying material slowly surrounds the bubble. In the present study, the gasbubbles are thus only ''held in place'' by a smaller fraction of the engulfing material. This causes the holding force, and consequently the internal pressure of the bubble, to be lower than in Han's research and explains why the ''eruption'' in the present experiments is not as explosive and only displaces the bubble by about 50 lm, while Han observed distances exceeding 1 mm.
A final reaction was observed in bubbles throughout the whole sample, but especially for bubbles that were close to the solidification front: popping of the bubble, as seen in Fig. 2 for t ¼ 84:6 s. Due to the aforementioned increasing growth rate of the bubbles with decreasing distance from the solidification front, small bubbles would grow bigger and pop at the top or bottom face of the sample.

Conclusions
Three major findings have been reported in this paper.
(i) It was shown that gas bubbles play a noticeable role during directional solidification of metals. The increased thermal gradient ahead of the bubble, resulting from the insulating properties of the gas pocket, directly impacts growth speed and homogeneity of the As can be partially inferred from the formula, each of these notations provides different advantages: ½H-atoms/g Alloy is a straightforward experimentcentered notation, ½ccm H 2 =100g Alloy STP is the most commonly used. As for ½ppmw HÀatoms , it correlates well with the commonly used ½%wt. for the elemental composition of the analyzed alloy. ½ppma HÀatoms , on the other hand, is the only notation that allows for the direct comparison of samples with different elemental compositions. This can be seen in Table 2 for the starred entries, where Al-33 %wt: Cu features a lower H-content than Al-4:5 %wt: Cu for the mass centered units, but when taking into account the actual number of atoms, the H-content in Al-33 %wt: Cu is higher.
Specifically, the ½ppm notations have great practical use, however, due to a lack of international norms addressing the exact nomenclature, a vast mix of variants are used by authors, sometimes without any clarification at all on whether atomic parts per million ½ppma, or weight parts per million ½ppmw are used, in some works even the syntax of the unit changing midways through the paper. Table 3 gives an overview over the most common notations for ½ppm.  [29] Values in the units used in the respective publication are marked in bold. The starred lines illustrate the problematic of mass-centered units displaying a higher H-atom content, despite there being a lower amount of H-atoms as shown by the ½ppma value