T. Pusztai, L. Rátkai, L. Horváth et al.
Acta Materialia 227 (2022) 117678
Fig. 1. Steady state solidification and melting profiles for a symmetric lamellar structure of lamellar spacing λ = 1.4λJH . The RGB components of the displayed colours are
the volume fractions of the phases: R = pβ , G = pL and B = pα . The 4 – 4 snapshots constituting panels (a) and (b) correspond to simulations with interface thickness
W = λJH /32, λJH /64, λJH /128, and λJH /256, from left to right. Note that the front position and shape are essentially independent of the choice of the interface thickness.
Fig. 2. Front temperature vs. lamella spacing during steady state solidification and
melting of lamellar structures for the symmetric model alloy. The symbols corre-
spond to the actual phase-field simulations in domains of different widths and with
different interface thickness values, while the solid lines show the prediction of the
Jackson-Hunt theory for solidification and its modified form (Eq. (13)) for melting.
The simulation results are well converged and in good agreement with the theory.
in the front position. To check the convergence of the simulations
with decreasing interface thickness and to explore how the inter-
face undercooling depends on the lamellar spacing λ around the
Jackson-Hunt value λJH predicted for solidification at the minimum
undercooling, we performed a series of simulation with domain
widths ranging from 0.6λJH to 1.7λJH and repeated this series with
interface thickness divided by 2, 4, and 8.
Fig. 1 shows the steady state solidification and melting struc-
tures for λ = 1.4λJH for the four values of the interface thickness.
As the interface thickness W and the pixel size x = 0.8W de-
creases simultaneously, the snapshots get “sharper”, but the shape
and position of the solid-liquid interface do not change, indicating
that we are working in a parameter range where the results do
not depend on the choice of the interface thickness. Note, however,
the differences between the front positions and the front shapes
observed in the cases of solidification and melting. Since only the
sign of the pulling speed was changed and the linear temperature
profile was kept fixed, the difference in lamella heights shows that
melting takes place at a higher temperature than solidification, as
expected. It is also observed that the interface curvature is the
lowest at the centre of the lamellae and the highest at the trijunc-
tion during solidification, whereas it is the lowest at the trijunction
and the highest at the centres of the lamellae during melting. This
finding is in agreement with the theoretical expectations presented
in Section 2.2.
Fig. 2 shows the interface temperature (defined as the temper-
atures at the lamellae tips, equal here for α and β ) as a function
of the lamella spacing λ. The simulation results are well converged
and they are in good agreement with the analytic predictions. It is
also confirmed that if a solid structure, which was prepared at the
minimum undercooling condition (λ = λJH ) corresponding to v, is
melted at a pulling speed −v, melting happens exactly at the eu-
tectic temperature.
In the following simulations, we start melting directly from the
initial structures (lamellar in 2D, rod in 3D) defined in Section 3.2.
Since these initial structures are good approximations of the eu-
tectic solid formed during freezing, no difference in the melting
results is expected. We made the comparison between the two ap-
proaches for a few cases, and found that indeed the melting pro-
cesses were indistinguishable.
So far we addressed fully symmetric cases: both the phase dia-
gram and the initial configurations were symmetric, yielding sym-
metric melting profiles. Next, we introduce asymmetry to the sys-
tem by using a non-symmetric solid phase ratio. We fixed the
lamella widths at 45% α and 55% β , which corresponds to c0 =
0.05, and we performed a series of simulations varying the lamel-
lar spacing λ between 0.7λJH to 1.7λJH . These simulations were re-
peated with temperature gradients increased by factors of 2 and 4.
The respective results are presented in Fig. 3. In this off-eutectic
setting, the α and β lamellae melt at different positions and tem-
peratures, and this position difference increases with the thermal
length. Apparently, lamellar spacing plays only a minor role in de-
termining the tip positions as in the symmetric case (see Fig. 1).
Note that the analytic theory outlined in Section 2, which assumes
a flat interface, is not applicable here.
Next, we clarify how the initial phase ratio affects the melting
geometry of the lamellar structure. We varied the lamella width
ratio, and thus c0 in a small range around the eutectic composition
and made simulations until steady state was reached. As shown in
Fig. 4, the phase whose amount was reduced relative the eutec-
tic case melts at the eutectic temperature TE , independently of c0 ,
while the other (of increased amount) melts at its liquidus tem-
perature. For the easier reference to these amounts of phases, in
the rest of this paper, we term them “sub-eutectic amount” and
“super-eutectic amount” (which are considered with respect to the
ratio corresponding to the eutectic composition).
Though in the cases presented so far stationary solutions were
established in the moving reference frame for long enough times,
in some cases such steady state could not be achieved.
One example is shown in Fig. 5. This simulation of an off-
eutectic system (c0 = 0.03) corresponds to a rather large (λ/λJH =
4) interlamellar spacing and relatively large (lT /ld = 1) temperature
gradient. In this simulation we observed oscillation near the tri-
junction. The oscillation was stable both in the sense that it did
not change its amplitude or frequency and that it did not disap-
pear when increasing the length of the simulation domain (check-