Diusion-Controlled Anisotropi

Growth of Stable and Metastable Crystal

Polymorphs in the Phase-Field Crystal Model
G. Tegze,1 L. Grnsy,1 G. I. Tth,2 F. Podmani zky,2 A. Jaatinen,3,4 T. Ala-Nissila,3,5 and T. Pusztai2
1
BCAST, Brunel University, Uxbridge, Middlesex, UB8 3PH, UK
2
Resear h Institute for Solid State Physi s and Opti s, H-1525 Budapest, P.O.Box 49, Hungary
3
Department of Applied Physi s, Helsinki University of Te hnology, P. O. Box 1100, FIN-02015 TKK, Finland
4
Department of Materials S ien e and Engineering,
Helsinki University of Te hnology, P.O. Box 6200, FIN-02015 TKK, Finland and
5
Department of Physi s, Brown University, Providen e, Rhode Island 02912
(Dated: May 16, 2009)
We use a simple density fun tional approa h on a diusional time s ale, to address freezing to the
body- entered ubi (b ), hexagonal lose pa ked (h p), fa e- entered ubi (f ), and simple ubi
(s ) stru tures in free growth. With appropriate model parameters, mimi king fa eted olloids, free
energy minimization predi ts stable b -uid oexisten e and metastable h p-uid, f -uid, and
s -uid oexisten es. The s stru ture is unstable. We observe diusion- ontrolled rystal growth
(of exponent 1/2) for the other polymorphs. The growth velo ity and its anisotropy are ompared
for b , h p, and f rystals forming in the same liquid. Finally, we show that varying the latti e
onstant of a s substrate, one may tune the epitaxially growing stru ture between b and f .
PACS numbers: 64.60.My,68.08.p,81.10.Aj,81.30.Fb

Depending on the onditions of freezing, under ooled eren e liquid, the length s ale, n = ( ref L )/L
ref

liquids may solidify to dierent rystal stru tures [1. the redu ed number density, BL = (L ref L kB T )
1
, and
The sele tion of the rystal stru ture is inuen ed by var- ref
BS = K(L kB T ) . Here L is the ompressibility
1
ious properties of the available rystalline phases in lud- of the liquid, K the bulk modulus of the rystal, while
ing the thermodynami driving for e of freezing, and the v = 1 emerges from Taylor expanding the ideal gas term.
kineti oe ient that des ribes the atta hment of atoms Adjusting v , one may in orporate the 0th order ontri-
to the surfa e of the rystal [1. An interesting question bution from three-body orrelations. This form of F an
that an only be answered using an atomisti approa h be dedu ed from the Ramakrishnan-Yussou type per-
is, how the growth rates of stable and metastable rys- turbative DFT [3 after some simpli ations [5, su h as
talline polymorphs forming in the same liquid do om- Taylor expanding the two-parti le dire t orrelation fun -
pare. Mole ular dynami s (MD) simulations have been tion in Fourier spa e up to 4th order. The resulting ee -
used widely to evaluate the growth anisotropy of the tive two-parti le dire t orrelation fun tion shows a single
solid-liquid interfa e [2. However, in su h simulations maximum, so the respe tive free energy prefers a given
the diusion- ontrolled regime annot be easily a essed, interparti le distan e. This approa h retains the ri h-
and the realization of a multipli ity of metastable and ness of the original DFT, and predi ts rystal stru ture,
stable phases is not without di ulties. Other promis- anisotropy, and elasti properties. This model is known
ing andidates are the mole ular theories of freezing that to rystallize to the b stru ture [5. We note, however,
rely on the density fun tional te hnique (DFT). In the that any periodi stru ture that honors this distan e shall
past de ades a variety of DFTs have been developed [3. be a lo al minimum of the free energy. Assigning free en-
However, simulation of rystal growth has be ame reality ergy to periodi stru tures su h as b , h p, f , and s ,
only re ently, within a dynami al extension of the DFT the PFC model is expe ted to predi t phase preferen e.
(DDFT) [4. These studies are restri ted to a few hun- Sin e the number density is a onserved quantity, the
dred atoms. Starting from a simplied free energy fun - time evolution of the system obeys onserved dynami s,
tional, a newly emerging DDFT-type approa h, known as
the phase-eld rystal (PFC) method [5, appears to be
  
n F
able to address rystalline freezing on a far larger s ale. = Mn , (2)
t n
In this Letter, we use the PFC method to ompare
the growth of several rystalline polymorphs in the same where a onstant mobility of the form Mn = [(1 +
under ooled liquid. The redu ed free energy is as follows. n0 )D /(kB T ref
L )] sets the time s ale for system evo-
lution, while n0 is the redu ed number density of the
vn3 n4
 
n
Z
F = dr 2 4
BL + BS (2 + ) n

+ , initial supersaturated liquid, D the self-diusion oe-
2 6 12 ient, and F /n is the rst fun tional derivative of the
(1) free energy with respe t to the eld n. An advantage of
where F = (F FLref )/(ref L 3
kB T ), FL
ref
and ref
L the PFC method is that it is able to address pro esses on
are the free energy and the number density of the ref- the diusive time s ale, whi h is by orders of magnitude
 2

0 1.2 to the most stable b phase is less than 10% [1(b).

-1 One of the most novel feature of the PFC model, on-
1
-2
trary to onventional phase eld models generally used to

 /
BCC

 x 104

-3 0.8
-4 address rystallization, that the anisotropy of the surfa e
0.6 energy, and also anisotropy of kineti origin is impli itly
-5
-6 BCC 0.4 in luded in the model. First, to gain information on the
-7 FCC FCC
anisotropy of the solid-liquid interfa ial free energy , we
HCP (a) 0.2 HCP (b)
-8 SC SC determined the stable and metastable equilibrium shapes
-9 0
-8 -6 -4 -2 0 -8 -6 -4 -2 0 (that minimize for a given luster volume the ontribu-
n0 x 102 n0 x 102 tion from and ree ts its anisotropy). Quasi equilib-
rium state was a hieved via growing from initial grains
FIG. 1: (Color online). (a) Driving for e of rystallization, of the respe tive rystal stru ture. Strongly fa eted do-
, and (b) /bcc vs the redu ed initial number density,
n0 , for the b , h p, f , and s stru tures. de ahedral, o tahedral and hexagonal-prism shapes were
observed for the b , f , and h p stru tures. This per-
fe t fa eting, (often seen in olloids [11) a result of a thin
( 1-2 mole ular layers) interfa e prevented us to evalu-
longer than the time a essible for other atomisti simu- ate the anisotropy for the f and b phases, sin e their
lations, su h as mole ular dynami s [5. Su h diusion- equilibrium shape were bounded ex lusively by {110},
ontrolled relaxation dynami s is relevant for olloidal and {111} fa es. For h p stru ture the bounding {1010}
systems and for deeply under ooled liquids, where the and {0001} fa es give 1010 /0001 = 1.08 0.01.
self-diusion of the parti les is expe ted to be the domi- Anisotropies of kineti origin were determined in pla-
nant way of density relaxation. For mole ular liquids at nar growth s enarios using domain length Lz = 1024
small under oolings, the hydrodynami mode of density in dimensionless unit (ee tive length of 512 be ause of
relaxation dominates. A ordingly, the present ompu- the periodi boundary onditions required by the spe tral
tations will be indi ative to the behavior of olloidal sys- numeri al s heme). We investigated the most important
tems. Indeed su h systems are known to rystallize to low index fa es, the {100}, {110} and {111} of the b
stru tures investigated here [6. and f , and the {0001}, {1010}, and {1120} of the h p
Eq. (2) has been solved numeri ally using a semi- stru ture. When tra king the interfa e, after a brief tran-
impli it Fourier spe tral method [7 using x = 1 and sient, the front position z displays a roughly z t1/2
= 1 dimensionless steps for spa e and time dis retiza- behavior [Fig. 2(e), indi ating a diusion- ontrolled
tion. As initial ondition at, n0 = 0.04 number density growth me hanism, often observed in olloidal systems
was used in the liquid domain (unless stated otherwise), [12. We have tted z = z0 + C(t t0 )1/2 paraboli fun -
and an analyti al approximation of the stable, bulk rys- tion to positiontime data, where nite size ee t was not
tal phases [8 in the rystalline domains. dete table. Here z0 is the initial position, C is the velo -
Model parameters were set to BS = 31/2 /2, B = ity oe ient, and t0 is the transient time. Analogously
BL BS = 5 105 , and v = 31/4 /2, aiming the to the linear kineti oe ient [2, the velo ity oef-
phase oexisten e regions are su iently lose to ea h  ient C dened as above is a measure of the interfa e
other to be able to make meaningful omparison of atta hment kineti s, and its dependen e on the rystallo-
the dynami s of ordering to b , f and h p latti e. graphi orientation des ribes the kineti anisotropy. The
The oexisten e regions were al ulated between the liq- velo ity oe ients at n0 = 0.04 are summarized in Ta-
uid and the rystalline phases (b , h p, f , and s ) ble I, data for f {111} fa e is missing, the rst layer on
via numeri al minimization of the free energy with re- our initial rystal was a sta king fault and ontinued as
spe t to the density distribution. For the spe ied a h p stru ture. In our investigation all interfa es were
model onstants, ommon tangent onstru tion yielded ideal rystal fa es, the near to lose-pa ked hexagonal
the following rystal-liquid oexisten e regions: b - planes grow far slower than the ones are more orrugated
liquid: 0.0862 < n0 < 0.0315, h p-liquid: 0.0865 < on the atomisti s ale. Note, that in our model system,
n0 < 0.0344, f -liquid: 0.0862 < n0 < 0.0347, the C values for dierent stru tures an be dire tly om-
and s -liquid: 0.0249 < n0 < 0.0216. For the shake pared, as they orrespond to essentially the same driving
of ompleteness we also al ulated the thermodynami s for e. We also nd that C in reases monotoni ally with
for the s latti e, although we were not able to a hieve n0 [Fig. 2(f), while the ratio of C values for dierent
metastable s ordering in our simulations. The driving orientations varies with n0 .
for e for rystalline aggregation (the grand potential den- In omparison to experiments, availability of rele-
sity dieren e relative to the initial liquid [9) is presented vant growth data is very limited. The velo ity ratio
in Fig. 1(a). It must be noted that not just the oexis- v100 /v110 2.3 for fa eted 3 He rystals (b ) [13, is lose
ten e regions are very similar for the b f , and h p- to the present 1.7 2.7, although in the experiment
liquid phases, but also the driving for e dieren e relative growth was not diusion ontrolled. The experimental
 3

ratio v1010 /v0001 2.8 observed for diusion- ontrolled 1.2 0

n ~
n BCC{100}
fa eted growth of al ium hydroxide (hexagonal but not 1
~
-1
~
0.8 n n BCC{110}
h p) [14, a ords reasonably with our 2.4 for h p. -2 ~
n BCC{111}
0.6
In ontrast, a wealth of data are available from MD -3

n x 102
0.4 -4
simulations for these stru tures: A relatively small ki-

n
0.2 -5
neti anisotropy has been indi ated for the b stru - 0
-6
ture, and the sequen e of the growth velo ities vary with -0.2 (a) (b)
-7
the applied potential [2, although usually 100 > 110 -0.4
-8
as here. The MD sequen e for the h p stru ture (Mg) 120 140 160 180 200 0 100 200 300 400
[2 agrees with the PFC result, however, the anisotropy z z
is smaller. Simulations for the f stru ture (Lennard- -7.1 1
Jones, Ni, Ag, Au, Cu, and Fe) [2 indi ate that the 0.8
-7.2
fastest and lowest growth rates apply for the {100}

n~min x 102
0.6
and {111} orientations. The ratio 100 /110 varies in

nmax
-7.3 0.4
the range of 1.4 1.8, as opposed to the PFC result
C100 /C110 = 0.97. These dieren es are attributable 0.2
-7.4
to the fa ts that the MD simulations refer to materials (c) 0 (d)
of low melting entropy (Sf kB ), whose rystal-liquid
-7.5 -0.2
interfa e extends to 4 5 atomi layers, whereas with 58 59 60 61 62 63 64 65 58 59 60 61 62 63 64 65
the present model parameters the PFC realizes a sharp
x 10-3
x 10-3
1.4
fa eted interfa e.

velocity coefficient (C)

160 BCC {100} BCC {100}
1.2
It is worth mentioning some interesting onsequen es
interface position
140 BCC {110} BCC {110}
BCC {111} BCC {111}
of the anisotropi growth. The bulk rystal densities 120 1
forming from the same liquid dier signi antly as rys- 100 0.8
80
tals growing along dierent fa es [Fig. 2(b), due to grow- 0.6
60
ing rystals realize bulk density of the maximum driving 0.4
40
for e relative to the liquid density at the interfa e (the 20 (e) 0.2 (f)
minimum (nmin ) in the depletion zone). Average densi- 0 0
ties n obtained by oarse graining via FIR ltering [X, 0 5 10 15 20 -5.5 -5 -4.5 -4 -3.5

x 10-3 n0 x 102
an example for b {100} fa e plotted against the lo-
al number density along the enterline in z dire tion FIG. 2: (Color online). B rystal growth as predi ted
on Fig. 2(a). A omparison for dierent b fa es is by the PFC model. (a) Lo al and ltered densities a ross
shown on Fig. 2(b). Most of the rystal fa es show- the liquid-b {100} interfa e. (b) Filtered densities for the
ing layer-wise growth via 2D nu leation, with nu leation {100}, {110}, and {111} interfa es. ( ) Depth of the depletion
frequen y spe i to the rystal fa e. Due to the time av- zone vs. time in (a). (d) Peak amplitude vs. time for (a). (e)
eraging inherent in DFT, we annot tell whether single Interfa e position vs. time and (f) the velo ity oe ients vs.
n0 for three low index orientations.
or multiple nu leation takes pla e. (Preliminary simula-
tions with Langevin noise imply multiple 2D nu leation,
although, adding noise to Eq. (2) is not free of on ep- atomi distan e of the bulk f stru ture (8.204) and the
tual di ulties [4.) Plotting nmin (t) versus time shows latti e onstant of the bulk b phase (9.021). With the
os illations in the layer-wise regime [Fig. 2( ), whi h latter hoi es the {100} fa e of s is ommensurable with
is a uniqe behaviour of diusive dynami s. The lling the {100} fa es of the bulk f and b stru tures, respe -
of a rystal plane is monitored by the amplitude of the tively. This s enario is analogous to depositing harged
respe tive density peak nmax that hanges in sigmoidal olloidal parti les on a square-patterned substrate and al-
steps [Fig. 2(d): The rst three weak steps stand for lows us to explore how stress inuen es the growth rate.
the density waves from liquid ordering at the interfa e,
followed by a large one representing the rystallization of
the layer, whi h is ompleted when the next liquid-side
TABLE I: Velo ity oe ient C for various interfa es of the
layer is almost fully rystalline. b , f , and h p stru tures at n0 = 0.04.
Finally, we investigate how phase sele tion is inu-
en ed by a foreign substrate of s stru ture represented Stru ture {100} {110} {111}
by a periodi potential term V (r)n added to the free en- b 0.824 0.002 0.474 0.005 0.948 0.003
ergy density. We use here a potential, whi h is V = f 0.916 0.002 0.948 0.002 -
0.0533 for spheri al regions of radius R0 lo ated on a {1010} {1120} {0001}
s latti e of latti e onstant a0 , and V = 0 otherwise. a0 h p 0.228 0.002 0.940 0.002 0.096 0.002
has been varied in a range that in orporates the inter-
 4

1.8
160 SIT33 proje t. A.J. a knowledges nan ial support from
FCC BCC FCC BCC
1.6
140 the Finnish A ademy of S ien e and Letters. T.P. is a
1.4
120 grantee of the Bolyai Jnos S holarship.
1.2
1 100
c/a

v-
0.8 80
0.6 60
0.4 40
0.2 (a) 20 (b) [1 D. M. Herla h, D. Holland-Moritz, and P. Galenko,
0 0 Metastable Solids from Under ooled Melts. (Elsevier, Am-
8 8.4
a0
8.8 9.2 8 8.4
a0
8.8 9.2 sterdam, 2007).
[2 H. E. A. Huitema, B. van Hengstun, and J. P. van der
Eerden, J. Chem. Phys. 111, 10248 (1999); J. J. Hoyt
FIG. 3: (Color online). Heteroepitaxy in the PFC: Axial ratio and M. Asta, Phys. Rev. B 65, 214106 (2002); D. Y.
c/a of the b t stru ture (solid line), and the average growth Sun, M. Asta, and J. J. Hoyt, Phys. Rev. B 69, 174103,
rate (dashed lines) vs. the latti e onstant a0 of the substrate. 024108 (2004); Z. G. Xia, D. Y. Sun, M. Asta, and J. J.
Hoyt, Phys. Rev. B 75, 012103 (2007).
[3 T. V. Ramakrishnan and M. Yussou, Phys. Rev. B 19,
The stru ture of the rystals that grew in our simu- 2775 (1979); W. A. Curtin and N. W. Ash roft, Phys.
lations on the {100} fa e of the s substrate, is body- Rev. A 32, 2909 (1985); J. F. Lutsko and G. Ni olis,
entered tetragonal (b t). The axial ratio c/a varies on- Phys. Rev. Lett. 96, 046102 (2006).
tinuously with a0 [Fig. 3(a), where c and a are the latti e [4 S. van Teeelen, C. N. Likos, and H. Lwen. Phys. Rev.
onstants of the b t stru ture perpendi ular and paral- Lett. 100, 108302 (2008).
[5 K. R. Elder, M. Katakowski, M. Haataja, and M. Grant,
lel to the surfa e of the substrate, respe tively. At the Phys. Rev. Lett. 88, 245701 (2002); K. R. Elder and M.
appropriate a0 values, we observe the f and b stru - Grant, Phys. Rev. E 70, 051605 (2004); K. R. Elder et
tures. These ndings suggest that phase sele tion an al., Phys. Rev. B 75, 064107 (207); J. Berry, K. R. Elder,
be ontrolled via the stru ture of the substrate. Indeed, and M. Grant, Phys. Rev. E 77, 061506 (2008).
ele trophoreti deposition of harged polymer olloids on [6 U. Gasser, E. R. Weeks, A. S hoeld, P. N. Pusey, and D.
at substrates with appropriate square-patterned holes A. Weitz, S ien e 292, 258 (2001); V. J. Anderson and
produ es both the f and the b stru tures [15. N. W. Lekkerkerker, Nature 416, 811 (2002); A. Yethiraj
and A. van Blaaderen, Nature 421, 513 (2003).
The average growth rate interpolates roughly linearly [7 G. Tegze et al., J. Comput. Phys. 228, 1612 (2009).
between the values for the f and b stru tures [Fig. [8 In the single-mode approximation, the redu ed number
3(a). While on the f side a planar growth front is densities for the rystals are: b : see K. A. Wu and
seen, for a0 > 8.6 a rough surfa e omposed of four- A. Karma, Phys. Rev. B 76, 184107 (2007), f : n =
sided pyramids bound by {110} fa ets evolves probably 8A{cos(qx) cos(qy) cos(qz)}, and s : n = 2A{cos(qx) +
as a result of oupled Mullins-Sekerka and Asaro-Tiller- cos(qy) + cos(qz)}. As the single-mode approximation
Grinfeld (MS/ATG) type instabilities, expe ted in the does not work for h p, we have used
a simple ansatz:
n =A{cos(2qy/ 3) + cos(qx
qy/ 3) cos(2/3
qx +
stress eld the substrate generates in heteroepitaxy [16. qy/ 3) + cos(qx +qy/ 3) cos(4/3 + qx + qy/ 3)

The instability is ree ted in the spatial variation of the cos(2/3 + 2qy/ 3)} cos[( 3/ 8)qz]. Here q = 2/a.
average growth rate. The enhan ed sensitivity to the The latti e onstant a and the amplitude A have been
instability on the b side is attributable to the lower determined by analyti minimization of the free energy.
interfa ial free energy relative to the f side [2. It is ev- [9 X = fX (nX ) fL /n(n0 )[nX n0 ] fL (n0 ), where
ident that the substrate plays a role (provides the stress), nX is the rystal density maximizing the driving for e.
as in ist absen e planar growth is observed. As expe ted [10 E. B. Sirota et al., Phys. Rev. Lett. 62, 1524 (1989).
[11 G. Y. Onoda, Phys. Rev. Lett. 55, 226 (1985); A. T.
for a MS/ATG type instability, it an be suppressed be- Skjeltorp, Phys. Rev. Lett. 58, 1444 (1987); D. V. Ta-
yond a lower and a higher riti al velo ity [16. In the lapin et al., Adv. Mater. 17, 1325 (2005); J. D. Wood-
ase of a0 = 8.83), these limits fall to n0 0.05 and ward et al., J. Phys. Chem. B 110, 19456 (2006).
n0 0.02. We also nd that with in reasing ross- [12 A. P. Gast and Y. Monovoukas, Nature 351, 553 (1991);
se tional area the roughness of the interfa e in reases. K. S htzel and B. J. A kerson, Phys. Rev. E 48, 3766
Work is underway to quantify these phenomena further. (1993); W. B. Russel et al., Langmuir 13, 3871 (1997);
This work has been supported by the EU FP7 Col- S. Derber et al., Physi a A 235, 204 (1997).
[13 V. Tsepelin et al., Phys. Rev. Lett. 88, 045302 (2002).
laborative Proje t ENSEMBLE under Grant Agreement [14 V. S. Harutyunyan et al., J. Mater. S i. 44, 962 (2009).
NMP4-SL-2008-213669 and by the Hungarian A ademy [15 N. V. Dziomkina, M. A. Hempenius, and J. G. Van so,
of S ien es under ontra t OTKA-K-62588. T.A-N. and Adv. Mater. 17, 237 (2005).
A.J. have been supported by the A ademy of Finland [16 I. Durand et al., Phys. Rev. Lett. 76, 3013 (1996).
via its COMP CoE grant, and by Tekes via its MA-