4.2.1.1 Qualitative observations of flow behaviour

Figure 9 shows the evolution of convective flow in the presence of dendrites in the primary mush using a shadowgraph. In the presence of dendritic solid (inset of figure 9a), compositional convection takes place in the form of plumes (figure 9a). Since the thermal and compositional diffusivities are considerably different (the ratio between mass and thermal diffusivities is of the order of 10⁻²), the composition of the plumes was preserved, while the plumes gained heat during upward motion. Non-uniformly distributed plumes and their uneven thermal interaction with the surroundings led to transport in the lateral direction driven by thermal convection (temperature measurement in the left and right sides at different horizontal planes is plotted in figure 10). The existence of the resulting horizontal thermal gradient was observed to be the primary reason behind the formation of DDLs (figure 9b).

Figure 9. Evolution of the DDLs during bottom-cooled solidification of ternary solutions (water–27 wt % NH₄Cl–5 wt % KNO₃): (a) local solutal convection and plume formation at 30 min; (b) the onset of DDLs at 70 min; (c) development of DDLs at 160 min; (d) disappearance of plumes at 720 min; (e) domination of localized solutal convection at 930 min; and (f) disappearance of DDLs at 1110 min.

Figure 10. Temperature plot during bottom-cooled solidification of water–27 wt % NH₄Cl–5 wt % KNO₃. The representative location of thermocouples along the left and right sides of the cell are shown in the inset (top right).

At the onset of DDLs in the present ternary system, the first $(R{a_{{S_{1\_Exp}}}} = g{\beta _1}\Delta {S_1}{h^3}/\nu {\kappa _{{S_1}}})$ and second $(R{a_{{S_{2\_Exp}}}} = g{\beta _2}\Delta {S_2}{h^3}/\nu {\kappa _{{S_2}}})$ compositional Rayleigh numbers were 1.5 × 10⁹ and 1.3 × 10⁹ (at 75 min), respectively, whereas the thermal Rayleigh number was 10.9 × 10⁶ (discussed in subsequent sections). The literature suggests that the onset of DDLs required a threshold composition convection (composition Rayleigh number = 1.5 × 10⁹) and thermal convection (thermal Rayleigh number = 13.6 × 10⁶) (Kumar et al. Reference Kumar, Srivastava and Karagadde2018a, Reference Kumar, Srivastava and Karagadde2019).

The composition of bulk liquid at the top of the cell was found to be water–20.5 wt % NH₄Cl–7.5 wt % KNO₃. Observations of the DDL in the central region of the cell are reported for the first time (figure 9b), whereas a DDL occurred from the top in binary cases (Kumar et al. Reference Kumar, Srivastava and Karagadde2018a). As the compositional convection (plume) increases, the localized compositional convection near the solid–liquid interface was suppressed, as shown in figure 9(c). Hence, a DDL was able to sustain near the solid–liquid interface region (figure 9c). This can be verified by the development of a temperature difference at 35 mm after 150 min, as shown in figure 10. With time, the plume disappeared (as the composition reached towards the cotectic), and this led to a decrease in the strength of the DDL (greyscale intensity of interface), as shown in figure 9(d). In the absence of plumes, localized compositional convection reappeared in the cell (figure 9e), which further disrupts the weak DDLs and homogenize the composition within the bulk. Beyond 1110 min (figure 9f), DDLs disappeared from the cell, and the composition of fluid near the solid–liquid interface was water–20.85 wt % NH₄Cl–9.8 wt % KNO₃, which is near the cotectic line on the phase diagram.

In summary, the formation of a DDL in the central regions of the cell can be attributed to relative density differences between the two rejected components. The prior studies of double-diffusive convection predominantly showed a DDL emerging from the top of the cell, due to the presence of only one solute component in the mixture (Kumar et al. Reference Kumar, Srivastava and Karagadde2018a). The presence of heavier component rejection shows a tendency of settling down, which appeared to disturb the later thermal gradients at the top of the cell and leads to suppress the formation of a DDL in the top region.

4.2.1.2 Quantification of the flow field

Different projections of the velocity field were captured during the solidification of water–27 wt % NH₄Cl–5 wt % KNO₃ (regime IIa) using PIV. The velocity contours in the front plane (x–y plane, $(\sqrt {{u^2} + {v^2}} )$) and the top plane (x–z plane, $(\sqrt {{u^2} + {w^2}} )$) are shown in figures 11(a,d) and 11(b,c,e,f), respectively. Figure 11(a) shows the plume velocity contours and vectors at 60 min, and, except for the plume, negligible velocity was observed in the bulk. The bulk fluid and the plumes were captured 20 mm below the free surface, as shown in figure 11(b). The free surface is at the highest position of liquid in the solidification cell.

Figure 11. (a,d) Front views (x–y plane): convective velocity contours and vectors of (a) plume and bulk fluid at 60 min and (d) DDLs. (b,c,e,f) Top views (x–z plane): convective velocity contours and vectors of (b) bulk fluid at 20 mm below from the top free surface, (c) free surface, (e) the layer in a DDL and (f) the interface of a DDL. The representative position of the laser sheet during the experiment for panels (b,c,e,f) are marked by the lines in panels (a,d).

The horizontal laser sheet was located 20 mm below the free surface of the liquid for easier identification of mixing and the location of plumes. The low-velocity regions at plume locations (plumes are marked by dashed circles in figure 11b) are due to the two-dimensional images captured at a plane perpendicular to the plume. Around the plume, an inward spiral circulation of flow was observed, indicating the plume enlargement flow phenomenon (figure 11b). At the free surface, the plume mixed with bulk fluid, and these convective patterns were observed in figure 11(c). The downward (y direction) velocity at the top of the cell is of the order of 10⁻² mm s⁻¹ (the bulk has very low downward velocity, bluish colour contour in figure 11a). However, the mixing between the bulk and the plumes at the free surface creates a horizontal velocity in the x (u) and z (w) directions. The magnitude of mixing flow is of the order of 10⁻¹ mm s⁻¹ as observed in figure 11(c).

The convective velocity in the DDL (figure 11d) was very low compared to the plume; hence, the plume and convective velocity in the layer were captured and analysed separately. The flow vectors within the DDLs (figure 11d) were obtained using 500 ms pulse width, to correctly capture the tracer particles. In high-pulse-width images, plume particles were not traceable in the interrogation window due to high velocity. For these reasons, the velocity in the DDL only can be quantified. The location of the plumes is marked in figure 11(d). The convective rolls in the layer of each DDL were separated by a low-velocity zone (interface of DDL). The convective roll direction and number of rolls in the DDL were decided by the plume. In figure 11(d), one plume was observed in the field of view, and it behaved like a cold plate; hence rolls flow in a downward direction near the plume in each DDL. The top view of the DDLs is shown in figure 11(e) (layer of DDL) and figure 11(f) (interface of DDL).

4.2.1.3 Scale analysis

From the literature (Kumar et al. Reference Kumar, Srivastava and Karagadde2018a), the length scale for salt-finger cells (d) under composition-driven binary alloy solidification is formulated as

(4.1)\begin{equation}d\sim 2\mathrm{{\rm \pi} }{\left( {\frac{{{D_S}\nu }}{{g\dfrac{{\Delta S}}{h}{\beta_S}}}} \right)^{1/4}},\end{equation}

where ${D_S}$ is the compositional diffusivity (m² s⁻¹), $\nu$ is the kinematic viscosity (m² s⁻¹), g is the gravitational acceleration (m s⁻²), h is the height of the liquid, ${\beta _S}$ is the compositional expansion coefficient and $\Delta S$ is the composition difference between plume and bulk composition at the onset of convection (wt %). In the binary mixture, the scaled diameter of salt finger or plume (d from (4.1)) is inversely proportional to${({\beta _S}\Delta S)^{1/4}}$. However, in ternary systems, the proportionality elements are not known in cases where two different compositional expansion coefficients and composition differences are experienced during the convection.

For studying such systems, a linear stability analysis was performed to validate the present experimental studies. Using the governing equations for mass, momentum, temperature and compositions in linear stability analysis, a characteristic equation has been obtained as follows (the detailed analysis is reported in the supplementary material, available at https://doi.org/10.1017/jfm.2021.1):

(4.2a)\begin{equation}\frac{{{k^2}}}{\sigma }(q + \sigma {k^2}) ={-} \frac{{R{a_T}{{\rm \pi} ^2}{a^2}}}{{q + {k^2}}} + \frac{{R{a_{{S_1}}}{{\rm \pi} ^2}{a^2}}}{{q + {\tau _1}{k^2}}} + \frac{{R{a_{{S_2}}}{{\rm \pi} ^2}{a^2}}}{{q + {\tau _2}{k^2}}},\end{equation}

where ${k^2} = {{\rm \pi} ^2}({a^2} + {n^2})$, ${\rm \pi} a$ and ${\rm \pi} n$ are the wavenumbers in the horizontal and vertical directions, and q is the perturbation rate coefficient. Other variables and non-dimensional parameters are:

(4.2b)\begin{equation}\left. \begin{gathered} {\textrm{Prandtl}\;\textrm{number}\quad \sigma = \nu /\kappa ,}\\ {\textrm{Lewis}\;\textrm{number}\quad {\tau_1} = {\kappa_{{S_1}}}/\kappa ,{\tau_2} = {\kappa_{{S_2}}}/\kappa ,}\\ {\textrm{thermal}\;\textrm{Rayleigh}\;\textrm{number}\quad R{a_T} = \dfrac{{g\alpha \mathrm{\Delta }T{h^3}}}{{\nu \kappa }},}\\ {\textrm{compositional}\;\textrm{Rayleigh}\;\textrm{number}\quad R{a_{{S_1}}} = \dfrac{{g{\beta_1}\mathrm{\Delta }{S_1}{h^3}}}{{\nu \kappa }},\quad R{a_{{S_2}}} = \dfrac{{g{\beta_2}\mathrm{\Delta }{S_2}{h^3}}}{{\nu \kappa }},} \end{gathered} \right\}\end{equation}

where ΔT and ΔS are the temperature and composition differences between the top and bottom planes, subscripts 1 and 2 correspond to the first and second components, and $\kappa ,{\kappa _{{S_1}}},{\kappa _{{S_2}}}$ are the kinematic diffusivity of temperature and compositions S ₁ and S ₂, respectively. The expansion coefficients of the temperature and compositions S ₁ and S ₂ are indicated by $\alpha ,{\beta _1},{\beta _2}$.

Using marginal stability analysis at the initiation of convection (q = 0) and in the absence of thermal convection, (4.2a) can be simplified to

(4.3)\begin{equation}{\tau _2}R{a_{{S_1}}} + {\tau _1}R{a_{{S_2}}} = \frac{{{\tau _1}{\tau _2}{k^6}}}{{{{\rm \pi} ^2}{a^2}}}.\end{equation}

Based on the Lewis number, (4.3) can be interpreted in two cases.

Case I: The Lewis numbers are different $({\tau _1} \ne {\tau _2})$ where these relations reach a minimum at ${a^2} = 0.5$ and ${n^2} = 1$ in (4.3) and it is simplified as

(4.4)\begin{equation}\frac{{g{h^3}}}{\nu }\left( {\frac{{{\beta_1}\Delta {S_1}}}{{{\kappa_{{S_1}}}}} + \frac{{{\beta_2}\Delta {S_2}}}{{{\kappa_{{S_2}}}}}} \right) = 657.\end{equation}

From this, it can be concluded that the effective compositional Rayleigh number in the ternary system is proportional to $({\beta _1}\Delta {S_1}/{\kappa _{{S_1}}} + {\beta _2}\Delta {S_2}/{\kappa _{{S_2}}})$.

Case II: Both components (S ₁, S ₂) have the same Lewis number $({\tau _1} = {\tau _2} = \tau )$, and (4.3) can be written as

(4.5)\begin{equation}\left. \begin{gathered} {R{a_{{S_1}}} + R{a_{{S_2}}} = 6.75\tau {{\rm \pi}^4},}\\ {\dfrac{{g{h^3}}}{\nu }\left( {\dfrac{{{\beta_1}\Delta {S_1} + {\beta_2}\Delta {S_2}}}{{{\kappa_S}}}} \right) = 657.} \end{gathered} \right\}\end{equation}

This analysis predicts that the effective compositional Rayleigh number in the ternary system is proportional to $({\beta _1}\Delta {S_1} + {\beta _2}\Delta {S_2})/{\kappa _S}$.

In the ternary system, the scale of a salt finger can hence be evaluated, and this can be formulated as (4.6) and (4.7) below with the help of (4.1), (4.4) and (4.5).

Case I: Lewis numbers $({\tau _1} \ne {\tau _2})$ are different,

(4.6)\begin{equation}d\sim 2\mathrm{{\rm \pi} }{\left( {\frac{{\nu h}}{g}\left( {\frac{{{\kappa_{{S_1}}}}}{{{\beta_1}\Delta {S_1}}} + \frac{{{\kappa_{{S_2}}}}}{{{\beta_2}\Delta {S_2}}}} \right)} \right)^{1/4}}.\end{equation}

Case II: Lewis numbers $({\tau _1} = {\tau _2})$ are the same,

(4.7)\begin{equation}d\sim 2\mathrm{{\rm \pi} }{\left( {\frac{{\nu h}}{g}\frac{{{\kappa_S}}}{{{\beta_1}\Delta {S_1} + {\beta_2}\Delta {S_2}}}} \right)^{1/4}}.\end{equation}

If the compositional expansion coefficient of the first component is very large compared with that of the other $({\beta _1} \gg {\beta _2})$, then the diameter of the salt finger can be estimated to be only driven by the second component.

For the present study, we assumed that the Lewis numbers of both compositions are the same, as the orders of the mass diffusion coefficients are similar in order (Tanaka Reference Tanaka1975). The analytical plume diameter is approximately 1.8 mm using (4.7), which compares closely with the average experimental plume diameter (calculated using an averaged pixel count) of 1.7 mm, as shown in figure 12. Figure 12(a) shows the plume location from the top view, and the path of the plume in the porous zone is shown in figure 12(b).

Figure 12. Location of the plumes from (a) the top and (b) the front views at 100 min during solidification of water–25 wt % NH₄Cl–5 wt % KNO₃. A zoomed-in view of the plume from the top view is shown in the inset of panel (a).

Similarly, a scale for the solutal convective velocity of the plume is evaluated from (Patterson & Imberger Reference Patterson and Imberger1980; Thompson & Szekely Reference Thompson and Szekely1988)

(4.8)\begin{equation}{V_{plume}}\sim \frac{{{{(R{a_{{S_{1\_Exp}}}} + R{a_{{S_{2\_Exp}}}})}^{0.5}}\tau \kappa }}{{{h_l}}},\end{equation}

where $R{a_{{S_{1\_Exp}}}} = g{\beta _1}\Delta {S_1}{h^3}/\nu {\kappa _{{S_1}}} = 1.5 \times {10^9}$, $R{a_{{S_{2\_Exp}}}} = g{\beta _2}\Delta {S_2}{h^3}/\nu {\kappa _{{S_2}}} = 1.3 \times {10^9}$ (at 75 min from the experiments). From (4.8), the analytical plume velocity was found to be 3.4 mm s⁻¹, and the experimental velocity, as measured by the PIV technique, was 1.45 mm s⁻¹. Since the present PIV experiments are two-dimensional in nature, the actual plume velocity could not be captured, as the plumes were wavy. The helical angle of the plume was estimated using the shadowgraphs and was used to obtain an approximate plume velocity, and it was found to be 3.6–3.9 mm s⁻¹, which is comparable to analytical results.

The literature suggests that, at the onset of a DDL, the maximum number of plumes were observed in binary systems (Chen Reference Chen1997; Kumar et al. Reference Kumar, Srivastava and Karagadde2018a). In contrast, this was not observed in ternary situations (figure 9b,c). At the onset of DDLs in the ternary system, the thermal Rayleigh number in a DDL $(R{a_{T\_DDL}} = 10.9 \times {10^6})$ was of the same order as in the binary system (13.6 × 10⁶), which is required for the onset of a DDL (Kumar et al. Reference Kumar, Srivastava and Karagadde2018a). (The thermal Rayleigh number for the DDL was calculated by the average temperature difference between the left and right sides at a horizontal plane $(\Delta {T_{horizontal}})$ and is formulated as $R{a_{T\_DDL}} = g\alpha \Delta {T_{horizontal}}{h^3}/\nu \kappa$ (Kumar et al. Reference Kumar, Srivastava and Karagadde2018a).) The compositional convection $(R{a_{{S_{_{1\_Exp}}}}} + R{a_{{S_{2\_Exp}}}})$ and plume velocity were higher in the ternary system, which led to an early onset of DDLs without the occurrence of a maximum number of plumes.

4.2.1.4 Effect of high compositions of NH₄Cl and KNO₃

As the initial composition of KNO₃ and NH₄Cl are increased, the distance between ‘P’ (equilibrium cotectic composition for ‘O’ initial composition) and ‘O’ increases (figure 1). This led to an increase in compositional convection $({\beta _1}\Delta {S_1} + {\beta _2}\Delta {S_2})$ in porous (primary mush and cotectic mush) zones and the bulk fluid. As a result, the plume diameter decreased (figure 13), and the number of plumes increased; this can be verified from (4.7). In experiments where the bottom plate temperature is lowered, the number of plumes can increase (not performed in the present work).

Figure 13. Locations of plumes from the top view for the initial compositions (a) water–10 wt % KNO₃–23 wt % NH₄Cl, (b) water–10 wt % KNO₃–25 wt % NH₄Cl, (c) water–15 wt % KNO₃–22 wt % NH₄Cl and (d) water–17 wt % KNO₃–23 wt % NH₄Cl. A zoomed-in view of the plume is shown for each case. The red rectangle in each main panel shows the location of the plume that is shown in the zoomed-in view.

In regime IIa, where the amount of KNO₃ is increased, this can lead to interesting flow behaviour. The representative area of these initial compositions is marked by a dotted circle in regime IIa on the ternary phase diagram in figure 14(a). During the solidification, KNO₃ was transported by the rejected liquid from primary mush (water + KNO₃ -enriched mixture) as well as cotectic mush (KNO₃ -enriched mixture). During the initial stages, due to the weaker nature of the plume, the bulk fluid was mostly stagnant. As plumes rise upwards, the bulk fluid moves downwards due to mass conservation. For the case of initial composition water–15 wt % KNO₃–23 wt  % NH₄Cl (figure 14b), the mass conservation can be represented through

(4.9)\begin{gather}\left. \begin{gathered} {{A_{plume}} \times {V_{plume}} \times \textrm{no}\textrm{.}\;\textrm{of}\;\textrm{plumes} = {A_{bulk}} \times {V_{bulk}},}\\ {\textrm{diameter}\;\textrm{of}\;\textrm{plume} = 1.12\;\textrm{mm},\quad {V_{plume}}\sim 3.6\;\textrm{mm}\;{\textrm{s}^{ - 1}}\textrm{,}\quad \textrm{no}\textrm{.}\;\textrm{of}\;\textrm{plumes} = 18,}\\ {{A_{bulk}} = 50 \times 70 - 18 \times \textrm{area}\;\textrm{of}\;\textrm{the}\;\textrm{plume},\quad {V_{bulk}}\sim 0.018\;\textrm{mm}\;{\textrm{s}^{ - 1}}.} \end{gathered} \right\}\end{gather}

The downward velocity of bulk was found to be of the order of 10⁻² mm s⁻¹, which was also observed from the velocity field from figure 11(a).

Figure 14. (a) The representative region with higher amount of KNO₃ in the initial composition is marked in regime IIa by the red ellipse. (b) Downward convective flow observed in the top section due to the occurrence of Rayleigh–Taylor instability in water–15 wt % KNO₃–23 wt % NH₄Cl and (c) the corresponding velocity contour (x–y plane).

In binary systems, the plume (less dense) mixes with the bulk (heavier than plume) at the top, leading to a decrease in the density of the bulk. However, in ternary or high-component systems, although the density of the fluid in the plumes is lower compared to the bulk, it can still have considerable amounts of the heaviest component (KNO₃, solutal expansion coefficient ~0.0065 wt %⁻¹). As indicative examples, plumes (composition, water–23 wt % KNO₃–4.7 wt % NH₄Cl, density 1172 kg m⁻³; or composition, water–24.9 wt % KNO₃–0 wt % NH₄Cl, density 1170.5 kg m⁻³) are typically ~0.02 % or 0.04 % lighter in density compared to that of the bulk (composition, water–16 wt % KNO₃–21 wt % NH₄Cl, density 1172.3 kg m⁻³; or composition, water–15 wt % KNO₃–23 wt % NH₄Cl, density 1171.05 kg m⁻³), even when the heavier component is present in the plume. However, after mixing at the top (composition, water–19.5 wt % KNO₃–12.85 wt % NH₄Cl, density 1172.6 kg m⁻³; or composition, water–19.95 wt % KNO₃–11.5 wt % NH₄Cl, density 1171.68 kg m⁻³), the liquid can be ~0.028 % or 0.05 % heavier, compared to that of the same bulk. As NH₄Cl is less in the plume, water + KNO₃ mixing with water + NH₄Cl + KNO₃ at the top can be heavier compared to the bulk. This implies that the plume (enriched in KNO₃) continuously mixes with the high -density bulk material, only at the top of the cell; the overall density of the layers of fluid at the top could become slightly heavier compared to that of the bulk. This peculiar case is likely to occur in ternary or higher component systems and not in binary. It will generally occur in a multi-component system if the components have large differences in their specific gravity.

With the progress in the solidification, due to increased density after mixing of the plume at the top, the previously accumulated (initial time of solidification) bulk fluid moves downwards via Rayleigh–Taylor instability. These motions can be identified by the downward motion of salt fingers at the top of the cell, and the region is marked by a dashed rectangular box in figure 14(b). Note that the temperature difference is negligible between the plume and the bulk liquid at the top of the cell. The thermal expansion coefficient (~10⁻⁴) is lower than the solutal expansion coefficient, which indicates the density change at the top is greatly influenced by composition.

During solidification of water–15 wt % KNO₃–23 wt % NH₄Cl, water–18.7 wt % KNO₃–16.4 wt  % NH₄Cl (1178.2 kg m⁻³) and water–18.3 wt % KNO₃–16.8 wt  % NH₄Cl (1176.8 kg m⁻³) were the compositions near the free surface of the cell and 15 mm below the free surface at 70 min, respectively. The downward flow can be observed with an average velocity of 0.2 mm s⁻¹ in the bulk fluid by PIV, as shown in figure 14(c). The increase in downward velocity of the bulk fluid (V_bulk) from order 10⁻² (the initial stage of solidification) to 10⁻¹ mm s⁻¹ occurred due to Rayleigh–Taylor instabilities. The convective motion induced by the Rayleigh–Taylor instability homogenized the temperature differences which were required for the onset of DDLs and hence delayed the onset of DDLs.

4.2.3.1 Formation of stagnant fluid layers

The primary mush induced random convection of the heavier water–NH₄Cl mixture and plumes originated from the cotectic mush, which transported low -density water -enriched mixture (figures 19b and 21a). The height of the random convective flow was restricted due to the accumulation of water -enriched mixture transported by the plume at the top of the cell. This region of random flow is termed the ‘random convective zone’. The accumulation region above this zone, up to the free surface of the cell, is termed the ‘stagnant zone’, as shown in figures 19(b) and 21(a). These two sections of bulk fluid can be observed by PIV. In the bulk fluid, higher convective velocity was observed in the random convective zone, and negligible velocity was observed in the stagnation zone, as shown in figure 21(b). With progress in solidification, the random convection was suppressed, leading to the homogenization of stagnant and random convective zones. Thus, the disappearance of separation of the bulk was observed eventually (figure 19c).

Figure 21. (a) Schematic for the mechanism of the separation in the bulk fluids. (b) Velocity contours and vectors in the bulk liquid where the stagnant zone has no velocity.

4.2.3.2 Shifting of facets to dendrites in the primary mush

The compositions of the stagnant and random convective zones were found to be water–21.2 wt % NH₄Cl–20.9 wt % KNO₃ and water–20.5 wt % NH₄Cl–23 wt % KNO₃, respectively, at 70 min. The composition in random convection was near to the cotectic composition. Owing to a high solid fraction (0.73) of growing facets in the primary mush, KNO₃ in the bulk depleted, causing a reduction in the growth of facets, thus suppressing random convection during the early stages of solidification (200 min). This resulted in increased NH₄Cl presence in the bulk fluid, shifting the composition regime towards regime IIIa. Beyond 200 min, the dendrites began to grow as primary solidifying component instead of facets, for the initial composition in regime IIIb, and local random convection was observed from the cotectic mush, as discussed in § 4.2.2 (for solidification in regime IIIa). The bulk fluid composition was measured to be water–22.2 wt % NH₄Cl–20.6 wt % KNO₃ at 220 min, which represents regime IIIa.

This hypothesis can be ascertained by observing the microstructures and shadowgraph images. At 300 min, the microscopic view shows the dendritic solid in the primary mush and a few facets ( previously the primary mush solid in figure 20a), as shown in figure 22(a). During the initial stages, the plume location was very small as it grew in the cotectic mush (figure 20b), but after the shift in the composition into regime IIIa, the plume location can be easily identified from the top view as shown in figure 22(b).

Figure 22. (a) After the shift of the bulk composition from regime IIIb to regime IIIa for initial composition water–20 wt % NH₄Cl–25 wt % KNO₃ (regime IIIb), at 300 min, this shows dendrites in the primary mush with already existing faceted microstructures in the primary mush (figure 20a). (b) Top view of plume locations after the shift of composition from regime IIIb to IIIa.

An interesting phenomenon of ‘regime shifting’ was observed in systems falling in regime IIIb. The high solid fraction due to faceted structure formation during the initial stages aided the homogenization of composition and temperature, leading to a composition shift to regime IIIa. However, regime IIIa experiences dendritic growth, which always tends to maintain a gradient in the bulk fluid. This is identified as the main reason for the system to remain in the regime with no change in the primary solid.

4.2.3.3 Temperature measurements

The temperature measurements also show an interesting phenomenon (figure 23). In the stagnant zone (90 and 100 mm height), a temperature difference was observed between the left and right sides of the cell, due to the presence of plumes (figure 19). A low-temperature gradient was observed in the random convective zone (60 and 80 mm height in figure 23) till 200 min owing to the homogenization by random convection. On the shift of composition (regime IIIb to IIIa), temperature differences appeared across the height of the bulk fluid.

Figure 23. Temperature plot during solidification of water–20 wt % NH₄Cl–25 wt % KNO₃.

In the presence of faceted growth, a gain in temperature (discussed in § 4.1.1 for faceted growth in figure 7; Kumar et al. Reference Kumar, Abhishek, Srivastava and Karagadde2020a) was observed at the end of random convection. However, in this regime, the gain in temperature disappeared, and flat temperature profiles were observed in the region; this is marked by a grey rectangular box in figure 23. The temperature rise in the liquid after the suppression of random convection was compensated by the exchange of heat with the plumes. A DDL formed and nominal cooling curves at all positions were observed henceforth, as shown in figure 23.