Transition Pathways Connecting Crystals and Quasicrystals
Jianyuan Yin1 , Kai Jiang2 , An-Chang Shi3 ,∗ Pingwen Zhang1 ,† and Lei Zhang4‡
1
arXiv:2007.15866v1 [cond-mat.mtrl-sci] 31 Jul 2020
School of Mathematical Sciences, Laboratory of Mathematics and
Applied Mathematics, Peking University, Beijing 100871, China.
2
School of Mathematics and Computational Science, Xiangtan University, Hunan 411105, China.
3
Department of Physics and Astronomy, McMaster University, Hamilton, Canada L8S 4M1.
4
Beijing International Center for Mathematical Research,
Center for Quantitative Biology, Peking University, Beijing 100871, China.
(Dated: December 1, 2021)
Transition pathways connecting crystalline and quasicrystalline phases are studied using an efficient numerical approach applied to a Landau free-energy functional. Specifically, minimum energy
paths connecting different local minima of the Lifshitz–Petrich model are obtained using the highindex saddle dynamics. Saddle points on these paths are identified as the critical nuclei of the 6-fold
crystals and 12-fold quasicrystals. The results reveal that phase transitions between the crystalline
and quasicrystalline phases could follow two possible pathways, corresponding to a one-stage phase
transition and a two-stage phase transition involving a metastable lamellar quasicrystalline state.
Since the discovery of quasicrystals characterized by
quasiperiodic positional order with non-classical rotational symmetries [1], tremendous progresses have been
made on the understanding of these fascinating materials
[2, 3]. Various quasicrystals have been reported [1, 4–7].
Besides examples from metallic alloys, quasicrystalline
order has been observed in different systems including
Faraday waves and soft matter [8–15]. While the structure of quasicrystals are now well understood [16], the
study of the thermodynamic stability of quasicrystals
and phase transitions between crystals and quasicrystals,
which requires the examination of the free-energy landscape of the system, remains a challenge. In particular,
the emergence of quasicrystalline order from crystalline
states, or the nucleation of quasicrystals, represents a
long-standing unsolved problem.
In general, nucleation of a stable phase from a
metastable state could be examined using three approaches, i.e. classical nucleation theory, atomistic theory, and density-functional theory [17, 18]. Within the
framework of the density-functional theory, the freeenergy landscape of the system is described by a freeenergy functional determined by the density of the molecular species. Stable and metastable phases of the system correspond to local minima of the free-energy landscape, whereas the minimum energy paths (MEPs) on
the free-energy landscape represent most probable transition pathways between different phases. Transition states
(i.e. index-1 saddle points) on the pathways can be identified as critical nuclei, representing critical states along
the transition pathways. This theoretical framework has
been applied successfully to various problems undergoing
phase transitions, including the nucleation and growth of
crystalline structures [19–21].
In this Letter, we examine the transition pathways
connecting quasicrystals and crystals within the framework of density-functional theory. Specifically, we develop an efficient numerical method using the high-index
saddle dynamics (HiSD) and apply it to a Landau freeenergy functional, i.e. the Lifshitz–Petrich (LP) model
[22], with local minima corresponding to two-dimensional
(2D) crystalline and quasicrystalline phases. MEPs connecting various local minima of the model are computed and critical nuclei of the 6-fold crystalline and
12-fold quasicrystalline states are identified. In particular, two MEPs connecting two ordered phases are obtained, revealing that the phase transitions between the
crystalline and quasicrystalline phases could follow two
possible pathways, corresponding to either a one-stage
phase transition or a two-stage phase transition involving a metastable intermediate quasicrystalline state.
Although our methodology applies to any free-energy
functional, we will focus on the LP model for simplicity. The LP model is a Landau-theory designed to explore quasicrystalline structures with two characteristic
length scales [22]. Despite its deceivingly simple form,
the LP model possesses a rich phase behaviour containing a number of equilibrium ordered phases with 2-, 6-,
and 12-fold symmetries [22, 23]. As such, this simple
Landau free-energy provides an ideal model system for
the study of transition pathways connecting crystals and
quasicrystals. The LP model assumes a scalar order parameter φ(r) corresponding to the density profile of the
molecules in a volume V . The free-energy functional of
the model is given by [22, 23],
Z
2
1
∇ 2 + 12 ∇ 2 + q 2 φ
F(φ) = dr
2
ε 2 α 3 1 4
− φ − φ + φ ,
(1)
2
3
4
where 1 and q are two characteristic wavelength scales.
The thermodynamic behaviour of this model is controlled
by two parameters, ε and α, where ε is a temperaturelike parameter and α is a parameter characterizing the
asymmetry of the order parameter. Possible equilibrium
phases of the model correspond to local minima of the
2
free-energy functional with the mass conservation, which
are solutions of the Euler-Lagrange equation of the system, δF
δφ = 0. The Euler-Lagrange equation has multiple
solutions, correspond to stable/metastable phases, transition states, and high-index saddle points of the model
system. The phase diagram of the LP model has been
examined by a number of researchers [22, 23].
The first step of the study is to find accurate solutions,
corresponding to crystals and quasicrystals, of the EulerLagrange equation for the free-energy functional Eq. (1).
Because quasicrystals do not have periodic order, special
numerical methods are needed to describe their structures accurately. In general, discretization methods for
quasiperiodic structures include the crystalline approximant method (CAM) [24] and the projection method
[25]. In this Letter, we adopt the CAM to approximate
quasiperiodic structures in the whole space with periodic
structures in a large domain with proper sizes.
For a given set of d base vectors {e∗1 , · · · , e∗d }, a reciprocal lattice vector k of d-dimensional quasicrystals can
be expressed as,
k = κ1 e∗1 + · · · + κd e∗d ,
κj ∈ R.
(2)
It is important to note that some of the coefficients κj
might be irrational numbers. A quasiperiodic function
φ(r) can be expanded as,
X
φ̂(k) exp(ik · r).
(3)
φ(r) =
k
Since some reciprocal lattice vectors cannot be represented as linear combinations of e∗i with integer coefficients, proper rational numbers L are chosen such that
Lκj of all the concerned reciprocal lattice vectors k could
be approximated as integers. As a result, a quasiperiodic
function could be approximated by a periodic function
with a period 2πL,
X
r
,
(4)
φ̂(k) exp ik ·
φ(r) =
L
k
where the reciprocal lattice vectors k are linear combinations of e∗j with integer coefficients. Within this approximation, the computational domain becomes [0, 2πL)d
with periodic boundary conditions for φ(r). For the 2D
π
. For
(d = 2) 12-fold quasicrystals, we have q = 2 cos 12
∗
∗
e1 = (1, 0) and√e2 = (0, 1), the coefficients to be approxπ
imated are 1, 23 , 21 , q cos 12
, q cos π4 , q cos 5π
12 . According
to the simultaneous Diophantine approximation [26], the
proper values of L are 30, 82, 112, 306, etc. We have
tested the accuracy of varying L and found that L > 112
gave results within the required accuracy. Therefore, we
set L = 112 or larger in our numerical calculations, and
use the spectral methods for Eq. (4) with N points in each
dimension to discretize the order parameter. The nonlinear terms in the Euler-Lagrange equation are treated
by using the pseudospectral method [24].
Multiple solutions of the Euler–Lagrange equation
= 0 are obtained. The simplest solution is φ(r) = 0,
corresponding to the homogeneous state (H). Beside this
trivial solution, a number of spatially inhomogeneous solutions, including the 6-fold crystalline state (C6) and
the 12-fold quasicrystalline state (QC), have been found.
Interestingly, a new lamellar quasicrystalline state (LQ)
that is periodic in one dimension and quasiperiodic in the
other dimension is identified. It is noted that a similar
structure was reported in [27]. The structures of these
three ordered phases are shown in real and reciprocal
spaces in Fig. 1. It is important to note that the Hessian
D2 F(φ) of these ordered states has different multiplicities of zero eigenvalues, corresponding to the numbers of
Goldstone modes of these states [28, 29].
While a stable state, corresponding to a local minimum of the free-energy functional, could be calculated
by gradient descent algorithms with proper initial configurations, finding a transition state is much more difficult because it does not correspond to a local minimum. Moreover, multiple zero eigenvalues of its Hessian
D2 F(φ) could lead to the degeneracy of transition states.
The problem is further complicated by the fact that there
is no a priori knowledge of the transition states. Most
of the existing methods for solving nonlinear equations,
such as homotopy methods [30–32] and deflation techniques [33, 34], are inefficient for this degenerate problem because of superabundant solutions from arbitrary
translation. The string method [35, 36] requires a good
initial path connecting the initial and final states. Because there is no obvious epitaxial relation between the
crystals and quasicrystals, a direct interpolation between
the initial and final states cannot lead to the MEP. Similarly, surface-walking methods for searching index-1 saddle points, such as gentlest ascent dynamics [37] and
dimer-type methods [38], are also not capable of computing the transition states of QC from metastable states
because the eigenvectors with zero eigenvalues for the
metastable states would be mistaken as the ascent direction, leading to a failure of climbing out of the basin.
In order to overcome these difficulties, we apply a numerical method based on the HiSD to compute the highindex saddle points corresponding to the transition states
of quasicrystals. The HiSD for index-k saddles (k-HiSD)
is governed by the following dynamic equation [39],
k
X
δF
(φ),
(5)
φ̇ = − I − 2
vj vj⊤
δφ
j=1
δF
δφ
where v1 , · · · , vk represent the ascent directions, which
are taken as the eigenvectors of the smallest k eigenvalues of the Hessian D2 F(φ). Specifically, for a metastable
state φ∗ whose Hessian D2 F(φ∗ ) has m zero eigenvalues, we use the locally optimal block preconditioned
conjugate gradient (LOBPCG) method [40] to calculate
{u∗1 , · · · , u∗m } as an orthonormal basis of the nullspace of
3
C6
LQ
QC
FIG. 1. Three stable ordered states in the √
LP model: C6 with f = F/V = −6.8 × 10−4 , LQ with f = −7.3 × 10−4 , and QC
−4
−6
with f = −7.5 × 10 at ε = 5 × 10 , α = 2/2. The computational domain is [0, 2πL]2 with L = 112 and a 20π × 20π square
is zoomed in for better illustration. Scale bars represent 10π. Each lower right inset is the diffraction pattern in the reciprocal
space and the arrows specify primitive reciprocal vectors (in Z).
the Hessian D2 F(φ∗ ) and u∗m+1 as a normalized eigenvector of the smallest positive eigenvalue. Next, we apply
the (m + 1)-HiSD by choosing φ(0) = φ∗ + ǫu∗m+1 as the
initial search position and vi = u∗i (i = 1, · · · , m + 1)
as the initial ascent directions for searching an index(m + 1) saddle. The small positive constant ǫ is used to
push the system away from the minimum, which could be
regarded as an upward search on a pathway map [41]. By
relaxing the (m + 1)-HiSD for time-dependent φ with updated ascent directions vi (i = 1, · · · , m+1) by LOBPCG,
a stationary solution φnew can be found, corresponding
to a degenerate transition state with only one negative
eigenvalue and m zero eigenvalues in most cases. If φnew
turns out to be a high-index saddle point, for instance, an
index-k saddle (k 6 m), we then implement (k − 1)-HiSD
to apply a downward search on a pathway map [41] to
search lower-index saddles until the degenerate transition
state is located. The MEP is then obtained by following
the gradient flow dynamics along positive and negative
unstable directions of the transition state.
FIG. 2. Transition pathway from H to QC with ε = −0.01,
α = 1. The QC critical nucleus shows a circular shape with
a small amplitude. Scale bars represent 10π.
First, we present the MEP connecting a homogeneous
phase to a quasicrystal. By choosing ε = −0.01 and
α = 1, QC has a lower free-energy density of f = F/V =
−2.7 × 10−3 than H with f = 0. The critical nucleus of
QC from H is shown in Fig. 2.It is composed of a clear
QC pattern at the centre of the nucleus surrounded by
damping density waves. The density wave at the centre
of the nucleus has a much smaller amplitude than that
of the corresponding QC state, indicating that the assumption of classical nucleation theory would be invalid
in this case. Along the transition pathway and beyond
the critical state, the nucleus grows isotropically with an
increasing amplitude at the centre, eventually reaching a
full QC phase (see Supplemental Movie 1). It is noted
that the nucleation and growth of quasicrystals from the
homogenous phase has also been simulated using a dy-
Next we demonstrate how a quasicrystal would transform to a crystal. We choose ε = 0.05, α = 1 so that
QC is a metastable state with f = −5.3 × 10−3 and
C6 is a stable state with f = −6.3 × 10−3 . For this
case we found two transition pathways connecting QC
to C6. In the one-stage transition pathway, a circular
critical nucleus of C6 shown in Fig. 3(a) is observed. Interestingly, the growing C6 nucleus after the critical state
shows that a transient state along the transition pathway
contains a new LQ interphase connecting the C6 and QC
states (see Supplemental Movie 2 for QC → C6 transition pathway). This finding indicates that the LQ could
serve as an intermediate state connecting the QC and
C6 phases. Indeed, a two-stage transition pathway from
QC to C6 has been obtained from our calculations. This
namic phase-field crystal model in [42]. The structure
of the critical nucleus and subsequent growing patterns
predicted in the current study are consistent with the
simulation results.
cri cal nucleus
f
1.0e-8
0.1
H
0
QC
0
1.2
-2.7e-3
-0.4
transi on pathway
4
two-stage pathway reveals a first transition from QC →
LQ and a second transition from LQ → C6 as shown in
Fig. 3(a). Nucleation at the first stage shows an ellipsoidal critical nucleus of LQ with the periodic direction
as the major axis. The energy barrier of the LQ nucleus
(∆f = 4.5 × 10−6 ) is lower than the energy barrier of the
C6 nucleus (∆f = 5.3 × 10−6 ), indicating that the QC →
LQ transition pathway is the more probable one. After
the QC → LQ transition, the second-stage transition follows the formation of another ellipsoidal critical nucleus
of C6 with the quasiperiodic direction of LQ as the major
axis and eventually to the C6 phase (see Supplemental
Movie 3 for QC → LQ → C6 transition pathway). It is
noted that two transition pathways have been observed
for the gyroid to lamellar transitions of block copolymers [21]. Furthermore, the appearance of a metastable
intermediate state as a precursor of the stable phase is
consistent with Ostwald’s step rule [43].
Finally we present results on the emergence of a
quasiperiodic structure from a periodic
structure. By
√
choosing ε = 5 × 10−6 and α = 2/2, C6 becomes a
metastable state with f = −6.8 × 10−4 and QC is a stable state with f = −7.5 × 10−4 . Again, two transition
pathways are computed in this case (see Fig. 3(b)). It
is noted that, along the one-stage transition pathway,
a larger computation domain with L = 306 was used to
avoid the effect of finite domain size on growth dynamics.
Similar to the phase transition from QC to C6, a circular
critical nucleus of QC is found on the one-stage transition
pathway. After nucleation, the size of the QC nucleus
increases with the appearance of the LQ interphase (see
Supplemental Movie 4 for C6 → QC transition pathway).
On the other hand, a two-stage transition pathway from
C6 to QC via a metastable LQ, as shown in Fig. 3(b),
has been obtained. The critical nucleus of LQ at the first
stage assumes an ellipsoidal shape with the quasiperiodic
direction as the major axis. The energy barrier of LQ nucleus (∆f = 1.7 × 10−6 ) is lower than that of C6 nucleus
(∆f = 4.6 × 10−6 ), indicating that the C6 → LQ transition would be more likely chosen than the direct C6 →
QC transition pathway. Because LQ and QC have similar
free energies in this case, a small driving force from LQ
to QC transition is expected. Therefore, a much larger
critical nucleus of QC is found. A full C6 → LQ → QC
transition pathway is shown in Supplemental Movie 5.
In summary, we developed an efficient numerical
method based on the HiSD to accurately compute critical nuclei and transition pathways between crystals and
quasicrystals. The computational challenge of the problem stems from the existence of multiple zero eigenvalues
of the Hessian of different ordered phases. We solved this
challenging problem by applying the HiSD to search for
high-index saddle points, resulting in the solutions corresponding to the degenerate critical nuclei. Application of
the numerical method to the Lifshitz–Petrich model reveals an interesting set of transition pathways connecting
crystalline and quasicrystalline phases.
For the transitions between the crystalline C6 and quasicrystalline QC phases, two transition pathways, corresponding to a one-stage direct transition and a two-stage
indirect transition, have been obtained. We found that a
new, one-dimensional quasicrystalline, LQ phase, which
shows periodicity in one direction and quasiperiodicity
in the other direction, plays a crucial role to connect C6
with QC. Compared with one-stage transition pathway
between C6 and QC, the two-stage transition pathway C6
↔ LQ ↔ QC is, consistent with the Ostwald’s step rule,
more probable because the LQ nucleus has a lower energy barrier. The current study reveals the mechanism of
the structural transition during quasicrystal nucleation.
The accurate numerical results provide a comprehensive
picture of critical nuclei and transition pathways between
periodic and quasiperiodic structures. Furthermore, the
proposed methodology is applicable to a wide range of
physical problems undergoing phase transitions.
This work was supported by the National Natural
Science Foundation of China (Grants No. 11861130351,
No. 21790340, No. 11421101, No. 11771368) and the Natural Science and Engineering Research Council (NSERC)
of Canada. J. Y. acknowledges the support from the Elite
Program of Computational and Applied Mathematics for
Ph.D. Candidates of Peking University.
∗
†
‡
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
shi@mcmaster.ca
pzhang@pku.edu.cn
zhangl@math.pku.edu.cn
D. Shechtman, I. Blech, D. Gratias, and J. W. Cahn,
Metallic phase with long-range orientational order and no
translational symmetry, Phys. Rev. Lett. 53, 1951 (1984).
C. Janot, Quasicrystals: A Primer (Clarendon Press,
Oxford, 1992).
J.-B. Suck, M. Schreiber, and P. Häussler, eds., Quasicrystals: An Introduction to Structure, Physical Properties and Applications (Springer, Berlin Heidelberg, 2002).
W. Steurer, Twenty years of structure research on
quasicrystals. Part I. Pentagonal, octagonal, decagonal
and dodecagonal quasicrystals, Z. Kristallogr. 219, 391
(2004).
A. P. Tsai, Icosahedral clusters, icosaheral order and stability of quasicrystals—a view of metallurgy, Sci. Technol.
Adv. Mater. 9, 013008 (2008).
S. Fischer, A. Exner, K. Zielske, J. Perlich, S. Deloudi,
W. Steurer, P. Lindner, and S. Förster, Colloidal quasicrystals with 12-fold and 18-fold diffraction symmetry,
Proc. Natl. Acad. Sci. U.S.A. 108, 1810 (2011).
M. Martinsons, M. Sandbrink, and M. Schmiedeberg,
Colloidal trajectories in two-dimensional light-induced
quasicrystals with 14-fold symmetry due to phasonic
drifts, Acta Physica Polonica A 126, 568 (2014).
W. S. Edwards and S. Fauve, Parametrically excited quasicrystalline surface waves, Phys. Rev. E 47, R788 (1993).
X. Zeng, G. Ungar, Y. Liu, V. Percec, A. E. Dulcey, and
J. K. Hobbs, Supramolecular dendritic liquid quasicrys-
5
(a) f
one-stage transi on
two-stage transi on
cri cal nucleus
transient state
Δf=5.3e-6
1.2
0.6
LQ
QC
0
-5.3e-3
LQ
Δf=4.5e-6
Δf=3.2e-6
C6
-5.8e-3
-6.3e-3
transi on pathway
(b) f
transient state
0.8
LQ
Δf=4.6e-6
0.4
C6
0
-6.8e-4
Δf=1.7e-6
LQ
Δf=9.3e-6
QC
-7.3e-4
-7.5e-4
transi on pathway
FIG. 3. Transition pathways between QC and C6. (a) from QC to C6; (b) from C6 to QC. One-stage transition pathway
(dot dash curve) includes a circular critical nucleus of QC and an inserted transient state after nucleation showing LQ serves
as an interface between QC and C6. Two-stage transition pathway (solid curve) includes two ellipsoidal critical nuclei and a
metastable LQ. Two critical nuclei are partially zoomed in for better illustration in (b). All critical nuclei are labeled by white
rings and all scale bars represent 10π.
tals, Nature 428, 157 (2004).
[10] K. Hayashida, T. Dotera, A. Takano, and Y. Matsushita, Polymeric quasicrystal: Mesoscopic quasicrystalline tiling in ABC star polymers, Phys. Rev. Lett. 98,
195502 (2007).
[11] D. V. Talapin, E. V. Shevchenko, M. I. Bodnarchuk,
X. Ye, J. Chen, and C. B. Murray, Quasicrystalline order in self-assembled binary nanoparticle superlattices,
Nature 461, 964 (2009).
[12] J. Zhang and F. S. Bates, Dodecagonal quasicrystalline
morphology in a poly(styrene-b-isoprene-b-styrene-bethylene oxide) tetrablock terpolymer, J. Am. Chem. Soc.
134, 7636 (2012).
[13] M. Huang, K. Yue, J. Wang, C.-H. Hsu, L. Wang, and
S. Z. D. Cheng, Frank–Kasper and related quasicrystal
spherical phases in macromolecules, Sci. China Chem.
61, 33 (2018).
[14] M. N. Holerca, D. Sahoo, B. E. Partridge, M. Peterca,
[15]
[16]
[17]
[18]
[19]
X. Zeng, G. Ungar, and V. Percec, Dendronized poly(2oxazoline) displays within only five monomer repeat units
liquid quasicrystal, a15 and σ Frank–Kasper phases, J.
Am. Chem. Soc. 140, 16941 (2018).
A. P. Lindsay, R. M. Lewis, B. Lee, A. J. Peterson, T. P.
Lodge, and F. S. Bates, A15, σ, and a quasicrystal: Access to complex particle packings via bidisperse diblock
copolymer blends, ACS Macro Lett. 9, 197 (2020).
W. Steurer and S. Deloudi, Crystallography of Quasicrystals: Concepts, Methods and Structures (Springer, Berlin,
Heidelberg, 2009).
D. Kashchiev, Nucleation: Basic Theory with Applications (Butterworth-Heinemann, Oxford, 2000).
J. W. Cahn and J. E. Hilliard, Free energy of a nonuniform system. iii. nucleation in a two-component incompressible fluid, J. Chem. Phys. 31, 688 (1959).
L. Zhang, L.-Q. Chen, and Q. Du, Morphology of critical
nuclei in solid-state phase transformations, Phys. Rev.
6
Lett. 98, 265703 (2007).
[20] R. Backofen and A. Voigt, A phase-field-crystal approach
to critical nuclei, J. Phys. Condens. Matter 22, 364104
(2010).
[21] X. Cheng, L. Lin, W. E, P. Zhang, and A.-C. Shi, Nucleation of ordered phases in block copolymers, Phys. Rev.
Lett. 104, 148301 (2010).
[22] R. Lifshitz and D. M. Petrich, Theoretical model for
Faraday waves with multiple-frequency forcing, Phys.
Rev. Lett. 79, 1261 (1997).
[23] K. Jiang, J. Tong, P. Zhang, and A.-C. Shi, Stability of
two-dimensional soft quasicrystals in systems with two
length scales, Phys. Rev. E 92, 042159 (2015).
[24] P. Zhang and X. Zhang, An efficient numerical method of
Landau–Brazovskii model, J. Comput. Phys. 227, 5859
(2008).
[25] K. Jiang and P. Zhang, Numerical methods for quasicrystals, J. Comput. Phys. 256, 428 (2014).
[26] H. Davenport and K. Mahler, Simultaneous Diophantine
approximation, Duke Math. J. 13, 105 (1946).
[27] Z. Jiang, S. Quan, N. Xu, L. He, and Y. Ni, Growth
modes of quasicrystals involving intermediate phases and
a multistep behavior studied by phase field crystal model,
Phys. Rev. Materials 4, 023403 (2020).
[28] J. Goldstone, A. Salam, and S. Weinberg, Broken symmetries, Phys. Rev. 127, 965 (1962).
[29] A.-C. Shi, Nature of anisotropic fluctuation modes in
ordered systems, J. Phys. Condens. Matter 11, 10183
(1999).
[30] C. Chen and Z. Xie, Search extension method for multiple
solutions of a nonlinear problem, Comput. Math. with
Appl. 47, 327 (2004).
[31] D. Mehta, Finding all the stationary points of a potentialenergy landscape via numerical polynomial-homotopycontinuation method, Phys. Rev. E 84, 025702 (2011).
[32] W. Hao, J. D. Hauenstein, B. Hu, and A. J. Sommese, A
bootstrapping approach for computing multiple solutions
of differential equations, J. Comput. Appl. Math. 258,
181 (2014).
[33] K. M. Brow and W. B. Gearhart, Deflation techniques for
the calculation of further solutions of a nonlinear system,
Numer. Math. 16, 334 (1971).
[34] P. E. Farrell, Á. Birkisson, and S. W. Funke, Deflation
techniques for finding distinct solutions of nonlinear partial differential equations, SIAM J. Sci. Comput. 37,
A2026 (2015).
[35] W. E, W. Ren, and E. Vanden-Eijnden, String method
for the study of rare events, Phys. Rev. B 66, 052301
(2002).
[36] Q. Du and L. Zhang, A constrained string method and its
numerical analysis, Commun. Math. Sci. 7, 1039 (2009).
[37] W. E and X. Zhou, The gentlest ascent dynamics, Nonlinearity 24, 1831 (2011).
[38] L. Zhang, Q. Du, and Z. Zheng, Optimization-based
shrinking dimer method for finding transition states,
SIAM J. Sci. Comput. 38, A528 (2016).
[39] J. Yin, L. Zhang, and P. Zhang, High-index optimizationbased shrinking dimer method for finding high-index saddle points, SIAM J. Sci. Comput. 41, A3576 (2019).
[40] A. V. Knyazev, Toward the optimal preconditioned eigensolver: Locally optimal block preconditioned conjugate
gradient method, SIAM J. Sci. Comput. 23, 517 (2001).
[41] J. Yin, Y. Wang, J. Z. Y. Chen, P. Zhang, and L. Zhang,
Construction of a pathway map on a complicated energy
landscape, Phys. Rev. Lett. 124, 090601 (2020).
[42] S. Tang, Z. Wang, J. Wang, K. Jiang, C. Liang, Y. Ma,
W. Liu, and Y. Du, An atomic scale study of twodimensional quasicrystal nucleation controlled by multiple length scale interactions, Soft Matter 16, 5718 (2020).
[43] W. Ostwald, Studies on the formation and change of solid
matter, Z. Phys. Chem. 22, 289 (1897).