Professional Documents
Culture Documents
a
Simulation in Materials Science Research Group, Advanced Institute of Materials Science, Ton Duc Thang University, Ho Chi Minh City, Viet Nam
b
Faculty of Electrical and Electronics Engineering, Ton Duc Thang University, Ho Chi Minh City, Viet Nam
c
Institute of Theoretical and Applied Research, Duy Tan University, Hanoi, 100000, Viet Nam
d
Department of Computational Physics, Hanoi University of Science and Technology, Viet Nam
Keywords: We report molecular dynamics simulation for SiO2.Al2O3 liquid. The result shows that the system contains small
microscopic segregation TO-subnets (T is Si or Al) and two large ones consisting of 18 - 35% of total atoms. The average size of SiO-
domain structure subnets is larger than AlO-subnets, and T-O-T angle distribution is quite different for three O types. The mi-
dynamics heterogeneity croscopic segregation occurs with formation of three distinct domains, whose number density is 57, 67 and 69
alumina-silicate liquid
atom/nm3. Diffusion proceeds by collective movements without and with exchange of coordinated atoms. The
molecular dynamics
exchange of coordinated atoms gives the major contribution. During 300 ps, there are Si-rich and Al-rich regions,
where immobile and mobile O resides respectively. By examining atom's flow through a fixed cubic lattice we
find that the system contains immobile and mobile regions where the atom flow significantly differs from each
other. The dynamic heterogeneity is related to non-uniform spatial distribution of bond breaking-reforming.
1. INTRODUCTION aluminum is very complicated when built into the tetrahedral network
[19,20,21,22,23]. At low alumina concentrations, Al atoms are four-
Amorphous and liquid silicates, a mixture of SiO2 with other oxides fold coordinated, but these AlO4 are accompanied with oxygen tri-
such as Al2O3, CaO, MgO, Na2O, are of fundamental importance for clusters, the structural units where an oxygen is bonded with three T (T
theoretical understanding of the glass transition and physical phe- is Si or Al). The existence of tri-clusters is confirmed by ab-initio mo-
nomena that one encounters in multi-component melts [1,2,3]. For lecular orbital calculation [12,13]. In contrast, the five-fold and six-fold
instance, the motion of sodium through specific channels is thought to coordinated Al atoms are dominant in the system with high alumina
cause very fast sodium mobility, and the existence of those channels is concentrations. It is suggested that AlO5 species dominated in high-
reflected by a pre-peak around 0,95−1 Å−1 in static structure factor [4]. temperature liquids, transform into AlO4 and AlO6 species in glassy
According to Angell's classification, silicate liquid is the strong type in state [21,23]. The study for silicate glasses up to 110 GPa [22] showed
opposite to fragile liquids which depart from Arrhenius viscosity-tem- that below 16 GPa the average T-O coordination number rapidly in-
perature relation [5]. Silicate in both liquid and glassy states comprises creases from 4 to 6, and then at higher pressures the structural change
a network of SiO4 tetrahedrons that are de-polymerized by forming in Al-O coordination from 6 to larger than 6 occurs, where Al atoms
non-bridging oxygen (NBO) as the modifier oxides are added. The de- play an important role in densification process. Molecular dynamics
gree of polymerization quantified via Qn species, where Q represents (MD) simulation has found the evidence for tri-clusters, micro-phase
SiO4 and n is the number of bridging oxygen (BO), dictates the viscosity separation and Al-rich network structure percolation through the tet-
and transport property. It is shown that the viscosity of de-polymerized rahedral Si-O network [17]. A strong correlation between the mean
melt increases with pressure, while the isothermal viscosity of poly- square displacement and the exchange of coordinated atoms is found,
merized melts shows anomalous pressure dependence [6,7]. and the T-O bond breaking-reforming takes place non-uniformly in
Alumina-silicates have been intensively studied by both experi- network structure [18]. In accordance to previous studies, the local
ments [8,9,10,11] and simulations [12,13,14,15,16,17,18]. Experi- environment around Al differs strongly from the one around Si. They
ments using different techniques such as NMR, IR and Raman spec- also show a strong distortion of SiO4 and the breakdown of inter-
troscopy, and X-ray scattering indicate that the chemical ordering of mediate range order with increasing alumina concentration. However,
⁎
Corresponding author: Phone: 0084985554433
E-mail addresses: tobavan@tdtu.edu.vn (T.B. Van), lethevinh@tdtu.edu.vn (L.T. Vinh).
https://doi.org/10.1016/j.jnoncrysol.2020.120457
Received 19 March 2020; Received in revised form 22 September 2020; Accepted 24 September 2020
0022-3093/ © 2020 Elsevier B.V. All rights reserved.
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
the relatively large scale fractures like the low-density, high-density inserted in the simulation box, where lSB equals the size of simulation
phase and microscopic segregation are not identified from those stu- box. The volume of G-polyhedron is given as VGP = mGVSB/m3, where
dies, although this has a specific meaning in the material science. Such mG is the number of nodes which are closer to G than other nodes; VSB is
issue is one among the aims of present simulation which is performed the volume of simulation box; VSB = lSB3. m is set to 220 and therefore
for the silicate liquid with high network modifier concentration. provides the high calculation accuracy. The volume of TO-subnet is
Dynamics heterogeneity (DH) could be extracted from multi-point given as the sum of polyhedron's volumes for all atoms of this subnet.
and multi-time correlation functions, which allow to finding the spatial Diffusion is mainly realized via T-O bond breaking-reforming
correlation between particle displacements for a finite time interval
[24,25,26,27]. Unlike phenomenological approaches, the atomic si-
TOx TOx+ 1 and OTy OTy + 1 (2)
mulation provides full access to trajectory of atoms, and one can get As the transformation (2) occurs, TOx and OTy exchange the co-
more details about the interplay between dynamics and local environ- ordinated atoms. To evaluate the atom's mobility, we calculate the
ment at atomic level in both space and time. The relationship between number of visited coordinated atoms nvca during the observation time
DH and structure heterogeneity for silicate liquids, as far as we know, tobs. As an example, Fig. 1b illustrates how to calculate nvca through four
remain poorly understood. These issues therefore stand in a focus of the consecutive configurations. Here the list of coordinated atoms changes
present work as well. Our calculation consists of several stages. Firstly, for different configurations. The configurations obtained from the si-
we build a large SiO2.Al2O3 model. Then we map out T-O subnets and mulation are separated by 10 ps. The nvca characterizes the mobility of
three O types. Next, structural and dynamics characteristics such as the given atom. To give more details about the dynamics, we specify two
Voronoi volume, T-O-T angle distribution and number of visited co- sets of mobile and immobile O, and denote them to SMO and SIMO,
ordinated atoms are calculated. Finally, the distribution of atoms respectively. Each set contains 10% of total O, i.e. 350 O, and the time
around nodes of a fixed cubic lattice and atom's flow through them are tobs is set to 300 ps. SMO and SIMO consist of atoms which have re-
determined. Our calculation is generic one and may be employed for spectively smaller and larger nvca than the one of the rest O. We also
other silicate structures. specify the third set of O (SRO) whose atoms are taken randomly from
the system. As we show below, because the rate of transformation (2)
2. COMPUTATIONAL METHOD for AlOx and OAly is significantly larger than that for SiOx and OSiy,
SIMO atoms are immobile and the majority of them are OSS and OSA. In
MD simulation is done in a cubic box containing 1000 Si, 3500 O contrast, SMO atoms are mobile, and most of them are OAA and OSA.
and 1000 Al under periodic boundary conditions. We use our self-de- The number <nvca> obtained by averaging nvca over all atoms of
veloped codes. The inter-atomic potentials of Born–Mayer type which SIMO, SMO and SRO equals 3.1, 73.38 and 37.29, respectively.
were successfully employed to simulate alumina silicate system (see e.g. To give a view about spatial distribution of atoms, we insert another
refs. [15,18,28,29]), are given as follows: cubic lattice with 10 × 10 × 10 nodes and length lSB. Then we de-
e2 r termine the number of atoms around each node. The atom is located
Uij (r ) = Zi Zj + Bij exp around a node if this node is closer to the given atom than other nodes.
r R (1)
Each node may be occupied or empty depending whether the atom is
with i, j = Al, Si, O. Here r is the interatomic distance and Zi, Zj are the located around it or not. Two nodes form a mutual linkage if they are
charges of ions: ZAl=3, ZSi=4 and ZO=-2. Other parameters of po- separated at the distance equal to the lattice constant. The node-cluster
tentials are: BAlAl=0 eV, BAlO=1779.86 eV, BOO=1500 eV, BSiSi=0 eV, consists of occupied nodes connected with each other by mutual lin-
BSiO=1729.5 eV and R=29 nm. The Verlet algorithm is employed to kages as illustrated schematically in Fig. 2. The spatial distribution of
calculate the atom motion with time step of 0.5 fs. The Coulomb in- atoms is identified by the distributions of atoms around node and size of
teraction is taken into account using Ewald-Hansen method. Initially, node-cluster.
we have prepared a well annealed sample at 5000 K, and then cooled DH can be evaluated by examining the atom flow through nodes of a
down it to 3000 K at constant volume. The sample is additionally re- fixed cubic lattice. To do this we insert a lattice with 12 × 12 × 12
laxed for 5 ns in the canonical ensemble (NVT). After that we performed nodes and length lSB, and then determine the list of atoms located
a long relaxation of about 30 ns in the micro-canonical ensemble (NVE) around each node during tobs. After that, we specify sets of immobile,
at temperature and pressure close to 3000 K and zero GPa, respectively. mobile and random nodes which are denoted to SIMN, SMN and SRN,
The obtained sample describes well the structure and dynamics of respectively. Each set contains 8% of total nodes. SMN, SIMN comprises
alumina-silicate liquid. More details about procedure to make the nodes whose navn is respectively larger and smaller than the one of the
model of alumina silicate liquid can be found elsewhere [14,15]. To rest nodes. Here navn is the number of atoms visiting the given node
collect the structural and dynamical data additional runs have been during tobs. SRN nodes are taken randomly from the lattice.
done to produce 100 consecutive configurations separated by 10 ps.
In order to calculate structural units (TOx and OTy) the cutoff dis-
3. RESULTS
tance rcutoff is taken from the position of first minimum of partial pair-
correlation function, rcutoff = 2.6 Å for Si-O and Al-O pair. The T-O bond
3.1. Structure of SiO2.Al2O3 liquid and microscopic segregation
is formed by T and O which stand from each other at the distance less
than rcutoff. All O are assorted into three types: OSA is connected with
Firstly, we examine the ability of potentials (1) to reproduce the
both Si and Al by T-O bonds; OSS and OAA are connected respectively
structural data of experiment and ab-initio simulation
with Si and Al atoms. Two types of TO-subnets are considered: i) SiO-
[29,30,31,32,33,34,35]. In general, the structural result obtained in our
subnet consists of Si and OSS which are bonded with each other by Si-O
simulation is close to the one from MD simulation [15] which, as shown
bonds, and ii) AlO-subnet comprises Al and OAA connected by Al-O
from studies [15,20], reproduces well the structure of alumina-silicate
bonds. The size of TO-subnet is defined as the number of atoms be-
liquid. Fig. 3 displays the partial and total pair-correlation functions.
longing to it. Different types of O and TO-subnets are illustrated in
Fig. 1a. The smallest TO-subnet contains one T with coordinated atoms For Si-O pair, the position of first peak takes the value of 1.58 Ǻ which
OSA. Thus, the structure consists of separate TO-subnets and OSA located is well consistent with experimental value of 1.58–1.6 Ǻ [32], but un-
in the boundaries between them. derestimates when compared to the result of ab-initio simulation [30].
Voronoi polyhedron has been calculated for every G atom (G- The function gSiO(r) remains essentially zero between 2.2-2.46 Ǻ in-
polyhedron). Here G is either of Si, Al or O. To calculate polyhedron's dicating the strong covalent Si-O bond. For pairs Al-O and O-O the
volume, a simple cubic lattice with m × m × m nodes and length lSB is position of first peak is located respectively at 1.66 and 2.64 Ǻ that
2
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
Fig. 1. a) OSS, OSA, OAA, SiO-subnet and AlO-subnet; b) Coordinated atoms for O labeled 11 in four consecutive configurations. Here the list of visited coordinated
atoms includes 2 Si and 4 Al; nvca = 6.
slightly underestimate the data of ab-initio simulation [30]. The gAlO(r) The fraction of structural units and three O types are listed in
is lower and broader with non-zero values between first and second Table 1. Among the range of coordinated atoms, SiO4 and AlO4 are
peaks indicating that the covalent Al-O bond is weaker than Si-O one. dominant (about 96% of total Si and 59% of total Al, respectively). The
The total pair-correlation function is shown in Fig. 3b. Here the com- fact that the fraction of OT2 remains relatively large but is markedly less
parison of our result is made with MD simulation [29] and previously than the unit indicating the modified tetrahedral structure of liquid. We
measured liquid at similar compositions [20]. The function g(r) re- also find a number of tri-clusters, and the fraction Cx = nOX/nO de-
produces well the shape and position of peaks, but its height over- creases in the order: SA → SS → AA. Probably, this trend takes place
estimate. The adding of dipolar expansion in potentials (1) could give due to much stronger Si-O bond compared to Al-O bond, and implies
better agreement with experimental g(r). We also compare the bond that SiO2 do not fully mix with Al2O3. The T-O-T angle distributions
angle distribution of our simulation. The peaks of O-Si-O, O-Al-O and shown in Fig. 4 are characteristic of the link between TOx units. The
Si-O-Si angle distribution at 105o, 105o and 160o, respectively, are in graph is quite different for three O types. In particular, the graph for OSS
reasonable agreement with the data of ab-initio simulation (109° for O- has a sharp peak at 160o indicating that Si forms the corner-sharing
Si-O angle, 107.6° for O-Al-O angle) and experiment (106.7° for O-Al-O tetrahedral network. Meanwhile, the distribution for OSA and OAA is
angle, 151.4° for Si-O-Si angle). Overall, the chosen potential provides a more complex, for instance, the distribution for OSA possesses a
small inconsistence for the pair-correlation function and bond angles, shoulder around 91o and a pronounced peak at 1530. According to
however this does not affect our major simulation results and conclu- previous studies [3,36,37], the oxygen tri-clusters occupying con-
sions siderable quantities are mainly composed of O (Al, Al, Al) and O (Al, Al,
3
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
Table 1
Fraction of structural units and three O types. Here CSiOx = nSiOx/nSi;
CAlOx = nAlOx/nAl; COTy = nOTy/nO; x and y are the number of coordinated O
and cation, respectively; CX = nOX/nO; nSiOx, nAlOx, nOTy and nOX are the number
of SiOx, AlOx, OTy and OX, respectively; nSi, nAl and nO are the total number of
Si, Al and O, respectively; OX is either OSS, OSA or OAA.
X CSiOx CAlOx y COTy X CX
4
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
Fig. 5. The number density (a) and volume fraction VD/VSB (b) as a function of time. Here VD and VSB are the volume of domain and total volume of simulation box,
respectively.
5
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
Table 3
MSD per atom for 10 ps. Here nI and nT are the number of first type atoms and
total atoms, respectively; <xI2> and <xII2> are the MSD for the first and the
second type atoms, respectively; <xT2> is the MSD for all atoms.
nI/nT <xI2> <xII2> <xT2>
sorted into two types. The number of visited coordinated atoms for the
first type is unchanged after 10 ps, while the one for the second type
increases. We observe that MSD for the second type atoms is sig-
nificantly larger than the one for the first type atoms, and MSD of the
first type atoms for Si, O and Al is close to each other. This again
confirms the fact that the transformation (2) gives the major con-
tribution to the diffusion. On the other hand, as <xT2> increases in the
order Si → O → Al, the diffusivity of Si is chiefly smaller than the one of
O and Al. The slowest diffusivity for Si is due to two factors: the fraction
nI/nT is larger and <xII2> is significantly smaller than those for O and
Al. This result follows that the diffusion is realized by two parallel
processes. On the one hand, the atoms move collectively without ex-
change of coordinated atoms and make small displacements. On the
other hand, they exchange the coordinated atoms via T-O bond
breaking-reforming and displace by large distances.
Fig. 8a presents the distribution of O around nodes for SMO, SIMO
and SRO. In general, the graph for SMO, SIMO strongly differs from that
for SRO. The number of SRO atoms around node is less than 4, while the
number of SMO, SIMO atoms around node varies in the range up to 6.
This result indicates that unlike SRO atoms which are uniformly dis-
tributed through the lattice, SMO and SIMO atoms tend to gather
around smaller number of nodes. Since all O atoms move from one to
other nodes, the atoms of specified sets will visit more nodes as the time
proceeds (see Fig. 9). The fraction of occupied nodes nON/nTN is shown
in Fig. 8b. SRO atoms spreads over whole lattice so fast that nON/nTN
practically reaches 1.0 after 500 ps. In contrast, nON/nTN for SIMO and
SMO approaches respectively to 0.347 and 0.735 indicating that the
atoms of those sets move not through whole volume, but only a sub-
volume of the model. In Fig. 8b we also plot nON/nTN for SIMO-SMO
nodes. The nON/nTN for SIMO-SMO nodes increases up to 0.15 which is
Fig. 6. a) The volume distribution for Si-polyhedrons, O-polyhedrons and Al- significantly smaller than the one for SIMO and SMO nodes. Therefore,
polyhedrons; b) The fraction of nodes versus the number of atoms around node. during the time tobs the system contains separate micro-regions where
Fig. 7. Fraction of unbroken T-O bonds (left) and MSD per atom versus nvca (right) as a function of time. Here MSD and nvca are determined for 300 ps.
6
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
4. CONCLUSION
7
T.B. Van, et al. Journal of Non-Crystalline Solids 552 (2021) 120457
Table 4 [9] I.A. Aksay, J.A. Pasic, R.F. Davis, J. Am. Ceram. Soc. 62 (7-8) (1979) 333–336.
Characteristics and size distribution of node-clusters. Here sNC is the size or [10] H. Morikawa, S. Miwa, M. Miyake, F. Marumo, T. Sata, J. Am. Ceram. Soc. 65 (2)
size's range of node-clusters; nSIMN, nSMN and nSRN are the number of node- (1982) 78–81.
[11] B.T. Poe, P.F. McMillan, C. Bertrand, M. Dominique, J.P. Coutures, J. Phys. Chem.
clusters for SIMN, SMN and SRN, respectively; <sSIMN>, <sSMN> and 96 (21) (1992) 8220–8224.
<sSRN> are the average size of node-clusters for SIMN, SMN and SRN, re- [12] B.T. Pole, P.F. Mcmillan, C.A. Angell, R.K. Sato, Chem. Geol. 96 (1992) 333–349.
spectively; <nASMN>, <nASIMN> and <nASRN> is the average number of [13] X. Xue, M. Kanzaki, J. Phys. Chem. B 103 (1999) 10816–10830.
atoms visiting SIMN, SMN and SRN node, respectively. [14] V.V. Hoang, Phys. Lett. A 368 (2007) 499–503.
[15] V.V. Hoang, Physica B, 400 (2007) 278-286.
sNC nSIMN nSMN nSRN Characteristics of node-clusters [16] P. Pfleiderer, J. Horbach, K. Binder, Chem. Geol. 229 (2006) 186–197.
[17] A. Winkler, J. Horbach, W. Kob, K. Binder, J. Chem. Phys 120 (1) (2004) 384–393.
1 3 2 90 <sSMN> 19.7 [18] L.T. Vinh, P.K. Hung, ToBa Van, N.V. Hong, Model. Simul. Mater. Sci. Eng. 28 (3)
2 1 1 4 <sSIMN> 23.0 (2020) 035008.
3 0 0 6 <sSRN> 1.3 [19] S. Krishnan, J.R.K. Weber, S. Ansell, A.D. Hixson, P.C. Nordine, J. Am. Ceram. Soc.
4-11 1 3 5 <nASMN> 116.2 83 (11) (2000) 2777–2780.
22-26 1 1 0 <nASIMN> 17.6 [20] M.C. Wilding, C.J. Benmore, J.K.R. Weber, J. Phys. Chem. B 114 (2010)
5742–5746.
96-106 1 1 0 <nASRN> 64.0
[21] M. Okuno, N. Zotov, M. Schmucker, H. Schneider, J. Non-Cryst. Sol. 351 (2005)
1032–1038.
[22] I. Ohira, Y. Kono, Y. Shibazaki, C. Kenney-Benson, A. Masuno, G. Shen, Geochem.
ACKNOWLEDGEMENT Persp. Let. 10 (2019) 41–45.
[23] R. Weber, S. Sen, R.E. Youngman, R.T. Hart, C.J. Benmore, J. Phys. Chem. B 112
(2008) 16726–16733.
The authors are grateful for support by the NAFOSTED Vietnam [24] K. Kim, S. Saito, J. Chem. Phys. 138 (2013) 12A506, 1-12.
(grant 103.01-2018.13). [25] G. Szamel, E. Flenner, Phys. Rev. E 74 (2006) 021507.
[26] N. Lacevic, F.W. Starr, T.B. Schroder, V.N. Novikov, S.C. Glotzer, Phys. Rev. E 66
(2002) 030101.
REFERENCES [27] C. Donati, S. Franz, S.C. Glotzer, G. Parisi, J. Non-Cryst. Solids 307–310 (2002)
215–224.
[1] D. Coslovich, G. Pastore, J. Phys.: Condens. Matt 21 (2009) 285107. [28] A. Winkler, J. Horbach, W. Kob, K. Binder, J. Chem. Phys 120 (1) (2004) 384–393.
[2] C. Le Losq, D.R. Neuville, P. Florian, G.S. Henderson, D. Massiot, Geochimica et [29] M. Bouhadja, N. Jakse, J. Phys.: Condens. Matter 32 (2020) 104002.
Cosmochimica Acta 126 (2014) 495–517. [30] J. Sarnthein, A. Pasquarello, R. Car, Phys. Rev. Lett 74 (1995) 4682.
[3] J.W. Drewitt, S. Jahn, C. Sanloup, C. de Grouchy, G. Garbarino, L. Hennet, J. Phys.: [31] M.T. Dove, D.A. Keen, A.C. Hannon, I.P. Swainson, Phys. Chem. Minerals 24 (1997)
Condens. Matt. 27 (2015) 105103. 311–317.
[4] A. Meyer, J. Horbach, W. Kob, F. Kargl, H. Schober, Phys. Rev. Lett. 9 (2) (2004) [32] R.D. Shannon, C.T. Prewitt, Acta Cryst. B25 (1969) 925.
027801. [33] M. Benoit, S. Ispas, M.E. Tuckerman, Phys. Rev. B 64 (2001) 224205.
[5] C.A. Angell, J. Phys. Chem. Solids 49 (8) (1988) 863–871. [34] F. Mauri, A. Pasquarello, G. Bernd, Pfrommer, Phys. Rev. B. 62 (2000) R4786(R).
[6] Y. Wang, T. Sakamaki, L.B. Skinner, Z. Jing, T. Yu, Y. Kono, C. Park, G. Shen, [35] N. Jakse, M. Bouhadja, J. Kozaily, J.W.E. Drewitt, L. Hennet, D.R. Neuville,
M.L Rivers, S.R. Sutton, Nat.Commun 5 (2014) 3241. H.E. Fischer, V. Cristiglio, A. Pasturel, Appl. Phys. Lett 101 (2012) 201903.
[7] N.K. Olga, N.A. Vsevolod, Anton Shatskiy, D.L. Konstantin, J. Non-Cryst. Sol. 375 [36] K. Zheng, Z. Zhang, F. Yang, S. Sridhar, ISIJ International 52 (3) (2012) 342–349.
(2013) 62–68. [37] Y. Xiang, D. Jincheng, M.S. Morten, J.C. Mauro, J. Chem. Phys. 139 (2013) 044507.
[8] R.F. Davis, J.A. Pask, J. Am. Ceram. Soc. 55 (10) (1972) 525–531.