Stability Properties of 1-Dimensional Hamiltonian Lattices With Nonanalytic Potentials
Stability Properties of 1-Dimensional Hamiltonian Lattices With Nonanalytic Potentials
Stability Properties of 1-Dimensional Hamiltonian Lattices With Nonanalytic Potentials
International Journal of Bifurcation and Chaos, Vol. 30, No. 15 (2020) 2030047 (19 pages)
c World Scientific Publishing Company
DOI: 10.1142/S0218127420300475
Thomas Oikonomou
Department of Physics, Nazarbayev University,
Qabanbay Batyr Ave 53, Nur-Sultan 010000, Kazakhstan
[email protected]
Bertin Many Manda‡ and Charalampos Skokos§
Department of Mathematics and Applied Mathematics,
University of Cape Town,
Rondebosch Cape Town, 7701, South Africa
‡[email protected]
§[email protected]
Received ; Revised
We investigate the local and global dynamics of two 1-Dimensional (1D) Hamiltonian lattices
whose inter-particle forces are derived from nonanalytic potentials. In particular, we study the
dynamics of a model governed by a “graphene-type” force law and one inspired by Hollomon’s
law describing “work-hardening” effects in certain elastic materials. Our main aim is to show
that, although similarities with the analytic case exist, some of the local and global stability
properties of nonanalytic potentials are very different than those encountered in systems with
polynomial interactions, as in the case of 1D Fermi–Pasta–Ulam–Tsingou (FPUT) lattices. Our
approach is to study the motion in the neighborhood of simple periodic orbits representing
continuations of normal modes of the corresponding linear system, as the number of particles N
and the total energy E are increased. We find that the graphene-type model is remarkably stable
up to escape energy levels where breakdown is expected, while the Hollomon lattice never breaks,
yet is unstable at low energies and only attains stability at energies where the harmonic force
becomes dominant. We suggest that, since our results hold for large N , it would be interesting
to study analogous phenomena in the continuum limit where 1D lattices become strings.
Keywords: Hamiltonian system; nonanalytic potential; simple periodic orbits; stable and unsta-
ble dynamics; local and global stability.
§
Author for correspondence
2030047-1
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
2030047-2
November 10, 2020 16:14 WSPC/S0218-1274 2030047
by nonlinear spring-mass equations of the form and discuss the two specific cases of graphene-type
mẍ = −Kx+ x|x|p−2 , 1 ≤ p < 2, for a single oscilla- and work-hardening interactions, providing theoret-
tor in the absence of external load. While for linear ical expressions for their periodic oscillations in the
elastic materials, the principal operator is the bi- single oscillator case. In Sec. 3, we consider the
Laplacian, for Hollomon’s power-law materials, it N particle case for both models and introduce a
is a bi-p-Laplacian [Wei & Liu, 2012; Wei et al., numerical stability criterion to identify the energy
2017b]. Here we plan to generalize these models by per particle E/N that corresponds to the first sta-
considering an array of N such coupled oscillators bility change of two of their SPOs, as E and N
described by the Hamiltonian increase. In Sec. 4, we study in more detail the
global dynamics of the graphene-type model, in the
N
1 vicinity of its SPOs after their first destabilization
H= mj ẋ2j
2 and use Lyapunov spectra to distinguish between
j=1
“weak” and “strong” chaos as the energy increases.
N
Finally, in Sec. 5, we conclude with a discussion of
K 2 λ μ
+ (xj+1 − xj ) + |xj+1 − xj | , (2) the results and an outlook for future research.
2 μ
j=0
2030047-3
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
For the graphene-type interactions we set C = Substituting this expression into the equation of
−D, D > 0 and q = 2, so that the Hamilto- motion (7), we find
nian takes the form of Eq. (1), while for the work-
(1 − ω 2 ) = |A1 ||cos ωt|, (10)
hardening interactions we have C = λ, λ > 0 and
q ∈ [0, 1) so that the Hamiltonian has the form which shows that ω < 1 as expected. In addition,
Eq. (2). We note that when q ≥ 1 the discontinu- |cos ωt| is a periodic function with period π/ω and
ity in the sign function when Δxj = xj − xj−1 = 0 can therefore be expanded as a Fourier series over
does not create difficulties regarding the numerical the interval [−π/2ω, π/2ω] as follows
integration since the term |xj − xj−1 |q dominates. ∞
However, when q ∈ [0, 1), which is the interval of B0
|cos ωt| = + Bi cos 2iωt, (11)
interest for Hollomon’s law, the sign function domi- 2
i=1
nates over the term |Δxj |q and creates spurious fluc-
tuations in the numerically computed total energy with
value, which should be constant. π/2ω
2ω 4
Thus, to avoid this undesired behavior in the B0 = dt|cos ωt| = ,
π −π/2ω π
numerical integrations, we approximate the sign
function in (4) by sgn(x − x0 ) ≈ tanh[τ (x − x0 )] π/2ω
2ω 4
for a value of τ > 0 large enough (typically τ = B1 = dt|cos ωt| cos 2ωt = , ....
π −π/2ω 3π
100). In what follows we will assume mj = 1, for
j = 1, . . . , N and fix the value of the exponent for (12)
the Hollomon-type interactions at q = 13 .
This implies
2 4
2.1. Graphene-type interactions 1 − ω 2 = |A1 | + cos 2ωt + · · · . (13)
π 3π
As explained in [Hazim et al., 2015; Wei et al., Therefore, we may find the frequency of these oscil-
2017a] and described above, a meaningful way to lations equating the constant terms
analyze a single graphene oscillator as a 1-degree
of freedom mass-spring system is through the 2|x0 |
ω2 = 1 − , (14)
equation π
setting A1 = x0 . In Fig. 1(a), we plot the sin-
mẍ = −Kx + Dx|x|, (7) gle cosine of Eq. (9) (blue curve) and the numer-
where m is the mass, K is the elastic coefficient and ical (black dots) solution over one time period for
D > 0 is a nonlinearity parameter. This equation is the initial amplitude x0 = 0.05 which corresponds
derived from the Hamiltonian function to a frequency ω 2 ≈ 0.968169 and find excellent
agreement.
m 2 K 2 D 3 For oscillations at higher energies, one has to
H= ẋ + x − |x| = E, (8)
2 2 3 consider higher harmonics of the Fourier series. For
instance, if we substitute into Eq. (7) the next
whose potential represents a symmetric well about
approximation of such a solution
x = 0, with extrema at x = ±K/D, where
the energy reaches its maximum value Hmax = x(t) = A1 cos ωt + A3 cos 3ωt, (15)
K 3 /6D2 . Thus, setting m = K = D = 1, and vary-
ing the energy we may study the periodic motions and expand it into Fourier terms
of the oscillator from small values E > 0 up to |A1 cos ωt + A3 cos 3ωt|
E = Emax = 1/6 beyond which the motion escapes
∞
to infinity and the mass-spring system “breaks”. B0
= + Bi cos 2iωt, (16)
Considering (with no loss of generality) the ini- 2
i=1
tial condition x(0) = x0 and ẋ(0) = 0, we may
approximate the low energy oscillation by a single we estimate the coefficients B0 , B1 in terms of A1 ,
harmonic term: A3 , and substitute them in the equation of motion
to obtain a nonlinear system of algebraic equations
x(t) = A1 cos ωt. (9) for ω 2 , A1 and A3 , for each initial amplitude x0 .
2030047-4
November 10, 2020 16:14 WSPC/S0218-1274 2030047
(a) (b)
Fig. 1. Numerical (black dots) versus analytical (blue curve) approximations of the single graphene oscillator (8) with unit
mass for (a) small (x0 = 0.05) and (b) large (x0 = 0.8) amplitudes of oscillations. In (a) we have used only the first term of
the Fourier series [see Eq. (9)] while in (b) we employed the first two terms [see Eq. (15)].
These low order approximations are quite accurate N = 2, for example, the potential has the form
even for x0 = 0.8 close to the separatrix, as we see in plotted in Fig. 2 and the escape energy threshold is
Fig. 1(b), where ω 2 ≈ 0.315843 and we compare the E esc
2 = 0.333 . . . .
analytical (blue curve) and numerical (black dots) It is, of course, highly desirable to estimate
results for the second order approximation (15). the escape energy thresholds of these models for
Proceeding now to higher dimensional gra- any N . To do this, one needs to find the criti-
phene-type models, with N = 2, 3, 4, . . . , it is easy cal points of V (x), solving the system of nonlin-
to see that the presence of the negative (abso- ear algebraic equations ∇V (x) = 0N , where V (x)
lute) values of cubic terms in the potential will represents the potential energy term in (3), with
always lead to escape at high enough energy. When x = (x1 , x2 , . . . , xN ). Then, one uses the Hessian
2030047-5
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
n 2
+ C1 Ai cos(iωn t)
i=1
n
− C2 ω 2n Ai cos(iωn t)
i=1
⎡ ⎤
n
×⎣ j 2 Aj cos(jωn t)⎦. (21)
j=1
Fig. 3. Logarithmic plot of the obtained approximations of
the escape energy thresholds per particle hesc esc
N = E N /N for
Using trigonometric identities to express the
the graphene-type model. squared quantities and the product in Eq. (21) as
2030047-6
November 10, 2020 16:14 WSPC/S0218-1274 2030047
(a) (b)
Fig. 4. (a) Plot of the numerical solution of the position and velocity functions in Eq. (18) with respect to time for the
rational exponent q = 1/3 and (b) Fourier series expansion for increasing n regarding the velocity function (black spheres)
at the vicinity of its minimum value within a period. As expected, for higher values of n we obtain a better approximation
solution.
2030047-7
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
Let us assume now that our N -dimensional lat- particles are located at the origin, and plot the pro-
tice possesses SPOs, which require that each mov- jections of orbits starting very close to a given SPO.
ing oscillator (some will be stationary) obeys the If the mode is stable, these projections will remain
same differential equation. This characterizes two very close to the SPO all time. However, at ener-
types of SPO solutions that will be of central impor- gies where the SPO has become unstable, nearby
tance in the remainder of this paper. They are orbits will start to move away from it, exploring a
continuations of the corresponding linear normal “chaotic” domain, whose size will give us informa-
modes of the λ = 0 case and have played a major tion about more “global” properties of the motion
role in similar studies of local and global stability around the SPO.
in FPUT lattices [Antonopoulos & Bountis, 2006; To determine the energy values at which these
Antonopoulos et al., 2006; Bountis & Skokos, 2012]. modes become unstable, we study the motion near
Among all possible nonlinear normal modes, these particles that are at rest in the exact periodic solu-
SPOs are the simplest ones, since all moving parti- tion, e.g. the second particle for the above SPO1
cles obey the same differential equation. Thus, they and the third particle for the SPO2. Varying the
involve the whole lattice in a uniform way and are total energy, we shift these particles by a distance
the easiest to study analytically and numerically. || 1 and calculate their maximum displacement
In the next section, we apply linear stability from zero as time evolves. Thus, we estimate the
analysis on two such SPOs of the nonlinear lattices energy of the first destabilization of the SPO when
described by Eq. (3) for the graphene and Hollomon this displacement becomes of the order of O(5||).
systems separately and compare the results. For example, in the case of the SPO1 mode with
N = 5, we select the initial conditions (ICs):
3. Stability of Simple Periodic x1 (0) = −x3 (0) = x5 (0) = x0 ,
(26)
Orbits x4 (0) = 0, x2 (0) = ,
As was done in the past for Hamiltonians with with x0 corresponding to the SPO’s ICs when the
analytic potentials [Antonopoulos & Bountis, 2006; system’s total energy is E, and study the dynamics
Antonopoulos et al., 2006; Bountis & Skokos, 2012], near this mode as E is changed. The same procedure
we also focus here on a pair of SPOs and investigate is applied e.g. to the SPO2 mode with N = 8, using
their stability. They are defined as follows: the ICs:
(1) SPO1 mode, N = 2k + 1, k = 1, 2, 3, . . .: x1 (0) = x2 (0) = −x4 (0) = −x5 (0)
x̂2j (t) = 0, x̂2j−1 (t) = −x̂2j+1 (t) ≡ x̂(t), = x7 (0) = x8 (0) = x0 , (27)
N −1 x6 (0) = 0, x3 (0) = ,
j ∈ 1, . . . , ,
2 choosing again x0 to correspond exactly to the
where every second particle is stationary SPO2 for energy E, and investigate how things
between two particles moving in opposite direc- change when E is varied.
tions. We have checked, of course, the accuracy of
(2) SPO2 mode, N = 3k + 5, k = 0, 1, 2, . . .: the above criterion against results obtained through
linear stability analysis, both for SPO1 and SPO2
x̂3j (t) = 0, x̂j (t) = −x̂j+1 (t) ≡ x̂(t), solutions, and have obtained very similar outcomes.
This demonstrates the reliability of our criterion
j ∈ {1, 4, 7, . . . , N − 1},
and allows us to bypass the time-consuming solu-
where every third particle is stationary, while tion of the so-called variational equations and the
the two in between move in opposite directions. computations of the monodromy matrix needed by
the linear stability analysis (for more details see
Among the q = 1, 2, . . . , N normal modes of the lin-
Appendix A).
ear lattice these SPOs are continuations of the ones
with q = (N +1)/2 and q = 2(N +1)/3 respectively.
To examine the motion in the vicinity of 3.1. Graphene-type interactions
these modes, we concentrate on a phase plane Let us apply the numerical approach described
(xi (t), ẋi (t)), i = 1, 2, . . . , N , where the stationary above to study the dynamics near the SPO1 mode of
2030047-8
November 10, 2020 16:14 WSPC/S0218-1274 2030047
our Hamiltonian (3) for q = 2, m = K = D = 1 and orbit, the solutions eventually wander over a much
C = −D describing N = 5 particles with graphene- larger chaotic region that spreads over most of the
type interactions. In Fig. 5 we present phase space available phase space of the system [Antonopou-
plots (xi (t), ẋi (t)), for i ∈ {1, 2, 3} for the first three los et al., 2006]. It is important to emphasize that
particles of the lattice at various energy levels, for entirely similar results are obtained when we con-
orbits with ICs of the form of Eq. (26) with x2 (0) = sider small displacements about the SPO2 orbit
0.01. Note that, at E = 0.17, the SPO1 mode is with N = 5.
still stable as the perturbed solution remains close Furthermore, if the motion near the unstable
to the periodic solution at distances comparable to SPO1 mode is chaotic, one would expect chaos
the initial displacement. At E = 0.21, however, the to be much “weaker” for orbits lying within the
SPO1 has certainly turned unstable, as the per- “figure-8” than those that spread over all of phase
turbed solution is oscillating at amplitudes that are space. Indeed, we have confirmed these expecta-
significantly larger than the initial ones. Finally, at tions by computing the corresponding Lyapunov
E = 0.22 chaos has clearly spread over all of the spectra (see Fig. 13 in Sec. 4) and verified that
available phase space, where the oscillations of all these two domains have truly distinct characteris-
particles become indistinguishable. tics: For the “figure-8” region, a single positive Lya-
What is remarkable here with regard to punov exponent is found and the remaining four
Fig. 5(b) is that, although the corresponding SPO1 converge to zero, while in the case of the larger
is clearly unstable, its nearby orbits remain within chaotic domain, four Lyapunov exponents are pos-
a limited domain surrounding this mode, and wan- itive and only one tends to zero, after sufficiently
der about it chaotically! This is highly reminiscent long integration times.
of similar results obtained for the FPUT lattice in Having thus tested the validity of our numer-
[Antonopoulos et al., 2006]. Indeed, as in the case ical stability criterion, we now employ it to deter-
of the FPUT 5 particle SPO1 mode, if we choose mine the first stability transitions of the SPO1 and
initial conditions very close to the unstable periodic SPO2 orbits of the graphene-type lattice as a func-
orbit, at energies where it has just become unstable, tion of the number of particles N . Earlier stud-
we observe that the chaotic orbits remain within ies on the FPUT lattice [Antonopoulos & Bountis,
a limited region shaped as a thin “figure-8” on a 2006] have shown that the destabilization energy
Poincaré surface of section (x1 , ẋ1 ) taken at times per particle hcN = E cN /N goes to zero by a power
when x3 = 0, as we see in Fig. 6. law as N increases, proportional to N −1 for the
Moreover, just as in the FPUT case, starting SPO1 case and N −2 for SPO2. As it turns out,
at points a little further away from the “figure-8” the situation for the graphene-type lattice is quite
2030047-9
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
Fig. 6. The “figure-8” chaotic orbit shown here arises near the SPO1 of an N = 5 graphene lattice for ICs at a distance
|| = 10−5 from the SPO. It clearly displays small-scale chaos, while, starting from ICs a little further away (|| = 10−2 ), the
orbits spread over much larger chaotic domains. The Poincaré surface of section (x1 , ẋ1 ) shown here is computed at times when
x3 = 0, with total energy E = 0.21, corresponding to what is shown in the middle plot of Fig. 5. The orbits were integrated
up to t = 2.5 × 104 .
different: Although the first destabilization energy, as periodic solutions of (3) with m = K = 1,
for both modes, falls to zero following a power law q = 13 and C = λ = 1.04. It is important to
hcN = E cN /N ∝ N −α , it does so with nearly the emphasize that, in all cases we tested, the results
same exponent α ≈ 1.72, as we can see in Fig. 7. described were found to be in very good agreement
with the predictions of local stability analysis (see
Appendix A).
3.2. Hollomon-type interactions Our aim is to determine in a similar way the
Let us turn now to our nonanalytic Hamiltonian critical energy per particle hcN = E cN /N at which
describing Hollomon-type interactions, and apply these fundamental modes change their stability.
the stability criterion described in the previous Remarkably, right from the start, we encounter a
subsections to study its SPO1 and SPO2 modes, surprising result, which is contrary to all other
Fig. 7. Logarithmic plot of the obtained approximations of destabilization energies per particle hcN = E cN /N for the graphene
lattice, superimposed with the power law ∝ N −1.72 .
2030047-10
November 10, 2020 16:14 WSPC/S0218-1274 2030047
(a) (b)
Fig. 8. Plot of the maximum oscillation amplitudes observed when applying small perturbations to the exact SPO1 solution
of: (a) a 7-particle graphene lattice and (b) a 7-particle Hollomon lattice with q = 13 . In (a) the destabilization at E c7 ≈ 0.218,
and in (b) the stabilization energy at E c7 ≈ 37.5 are indicated with a red line.
Hamiltonian lattices studied so far: The SPO1 and comparison purposes we show the corresponding
SPO2 modes are unstable at low energies and first stability transition for the SPO1 orbit of a N = 7
become stable at energy values that increase as N particle graphene lattice in Fig. 8(a).
increases! This becomes evident by applying our One possible explanation for this behavior is
numerical criterion to the SPO1 mode of a N = 7 the fact that the dynamics of the Hollomon lat-
particle Hollomon lattice, see Fig. 8(b). Starting tice, for small displacements, is governed by the
with small displacements, we find that the oscil- terms |xj+1 − xj |q+1 , which for 0 ≤ q < 1 can be
lations about this mode grow indicating instability, larger than the harmonic terms and may thus be
until the energy reaches a value E c7 ≈ 37.5. For responsible for the instability of the system at low
Fig. 9. Logarithmic plot of the approximate energies per particle hcN = E cN /N where the first stabilization of SPO1 and
SPO2 happens for the Hollomon lattice, showing a power law behavior of the form ∝ N β , with β ≈ 2.68 (dashed line).
2030047-11
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
Fig. 10. Logarithmic plot of the obtained approximations of SPO1 stabilization energies per particle hcN = E cN /N , for various
values of q ∈ { 13 , 35 , 57 } for the Hollomon lattice.
energies. As the energy grows, however, for fixed N , models. In the case of the graphene 1D lattice,
the harmonic terms in the potential become domi- after introducing numerical criteria to locate where
nant, which might explain why the motion becomes transitions happen, we noted that at energies just
stable and remains so at all energies above the sta- above destabilization, the motion near the SPOs
bilization threshold. does not wander over large distances in phase space,
Thus, our next task is to calculate the criti- but remains in a regime termed “weakly chaotic”
cal energy per particle hcN = E cN /N at which the in previous works on FPUT models [Antonopoulos
first transition to stability occurs. Estimating the et al., 2006; Bountis & Skokos, 2012]. It is only when
stabilization energies per particle for SPO1 and we start at further distances from the SPO that
SPO2 and plotting them in a double logarithmic the orbits begin to wander over a wider domain of
scale in Fig. 9, we find that they grow monotoni- phase space, exhibiting what we might call “strong
cally with N , both following nearly equal asymp- chaos”.
totic power laws of the form N 2.68 . In previous works [Antonopoulos et al., 2006;
Finally, it is interesting to investigate the effect Bountis & Skokos, 2012], a clear distinction was
of the exponent q = 2s−1 2s+1 , s ∈ N on the first stabi- made between “weak” and “strong” chaos by study-
lization energies of the SPOs. Results for s = 1, 2, 3 ing their spectra of Lyapunov exponents (LEs)
are presented in Fig. 10. The curves of SPO1 stabi- which differ significantly. Thus, in this section, we
lization energies per particle hcN for large enough N perform a similar investigation to reveal the global
are well fitted by power laws, so that hcN = αN β ⇒ dynamical properties of the SPOs of the graphene-
hcN ∝ N β . The approximation of the exponents are type Hamiltonian studied in Sec. 3.1 just after their
β = 2.68, 5.58, 9.61 for q = 13 , 35 , 57 , respectively. first destabilization.
Clearly, for low values of N , as q → 1, the stabi- As is well-known, the Lyapunov spectrum for
lization energies per particle become smaller, as the an orbit of an N -degree of freedom autonomous
effect of the harmonic terms begins to dominate at Hamiltonian system consists of 2N LEs λ(k) , k =
lower energies. However, as N increases, the behav- 1, 2, . . . , 2N , which measure the mean exponential
ior of the system tends to coincide for all the above rate of divergence (or convergence) of orbits in
choices of the exponent q. the immediate vicinity of the studied solution (see
[Benettin et al., 1976; Benettin et al., 1980a, 1980b;
4. Lyapunov Exponents and Global Skokos, 2010; Pikovski & Politi, 2016] and refer-
Stability for the Graphene Model ences therein). The LEs come in pairs of opposite
sign values
We have been interested so far in the first
de(re)stabilization of the SPOs of our nonanalytic λ(k) = −λ(2N −k+1) , k = 1, 2, . . . , 2N, (28)
2030047-12
November 10, 2020 16:14 WSPC/S0218-1274 2030047
2N (k) least twice differentiable, which is not the case for
so that k=1 λ = 0, with the largest N LEs
ordered as the Hamiltonians considered in this study. Thus, we
employ the so-called two-particle method [Benettin
λ(1) ≥ λ(2) ≥ · · · ≥ λ(N −1) ≥ λ(N ) = 0. (29) et al., 1976; Mei & Huang, 2018], which is based
The studied orbit is said to be chaotic if at least one on the simultaneous evolution of the studied orbit,
of its LEs is positive, which means that the maxi- along with several ones close by.
mum Lyapunov exponent (MLE) λ(1) > 0. On the For the numerical computation of the LEs,
other hand, if λ(1) = 0 the orbit is said to be regular. we evolve all required orbits by implementing
If, besides the MLE, more exponents are positive the SABA2 symplectic integrator (SI) of order 2
λ(k) > 0, k = 2, 3, . . . , k < N , then it follows [Laskar & Robutel, 2001]. Given a particular orbit,
that there are k directions (in an orthogonal refer- with ICs for x and ẋ at t = 0 denoted by X(0),
ence frame moving with the orbit), along which the we choose an appropriate number of nearby orbits,
(k)
motion is exponentially unstable. Thus, one might X̃(k) (0) at distance d0 = X(0) − X̃(k) (0) ≈
−8
argue that the higher the k the “more chaotic” is 10 , to keep the magnitude of the deviation vector
a given orbit, as more directions exist along which small and ensure the accurate evaluation of the LEs
nearby solutions can exponentially deviate away [Mei & Huang, 2018]. The phase space coordinates
from it. of these nearby orbits are randomly chosen from a
The values λ(k) of the LEs are obtained as the uniform distribution. All orbits are integrated up to
time limits the final time t = 105 with an integration time step
τ = 5 × 10−4 , which keeps the value of the relative
λ(k) = lim Λ(k) , (30)
t→∞ energy error
of appropriately computed quantities Λ(k) , usually H(t) − H(0)
Er (t) = , (31)
refereed to as the finite time LEs (ftLEs). These H(0)
quantities can, for example, be evaluated by the so-
called “standard method” (see e.g. [Benettin et al., smaller than 10−8 .
1980b; Skokos, 2010]). Typically this computation The computation of the MLE offers an alter-
is done through the numerical solution of the so- native way to investigate the stability changes of
called variational equations (see Appendix A for SPOs and corroborate the results in Sec. 3 (e.g.
more details), which govern the time evolution of Figs. 7 and 9), as λ(1) > 0 (λ(1) = 0) corresponds to
small perturbations from the studied orbit. A draw- an unstable (stable) periodic orbit. Typical exam-
back of this approach, however, is that it requires ples of these behaviors are shown in Fig. 11 where
the Hamiltonian function to be continuous, and at we present the ftMLE evolution for the SPO1 of
(a) (b)
Fig. 11. Time evolution (in log–log scale) of the ftMLE Λ(1) (t) of the SPO1 with energy (a) E = 0.1305 and (b) E = 0.2130,
respectively below and above the orbit’s first destabilization energy E c5 ≈ 0.21, for the Hamiltonian describing graphene-type
interactions. In both panels the dashed straight line corresponds to a function ∝ t−1 . The insets show the time evolution of
the relative energy error Er (t) of Eq. (31).
2030047-13
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
the graphene Hamiltonian model with N = 5 at orbit, and find how it varies as the system’s energy
energies E = 0.1305 [Fig. 11(a)] and E = 0.2130 and dimensionality change. Thus, to apply this to a
[Fig. 11(b)] respectively below and above the energy stable SPO we consider orbits whose ICs are located
E5c ≈ 0.21 of the SPO’s first destabilization. In further and further away in phase space from the
Fig. 11(a) we see that eventually Λ(1) ∝ 1/t, which SPO (the distance d between the two ICs is com-
is the typical asymptotic evolution of the ftMLE for puted as the usual Euclidean distance of points in
regular orbits (see e.g. [Skokos, 2010]), so that in the multidimensional spaces) and determine their regu-
large time limit λ(1) = 0. This behavior clearly indi- lar or chaotic nature. Then, the width of the regular
cates that the orbit is stable. On the other hand, region is quantified by the largest d value (denoted
for E = 0.2130 [Fig. 11(b)] the ftMLE converges by Dm ) for which the nearby orbit is regular.
towards a fixed positive value, which at time t = 105 In Fig. 12, we present results for the Dm of the
is λ(1) ≈ 0.014. This behavior suggests that the regular region around an SPO1 of the Hamiltonian
SPO1 is unstable. describing graphene-type interactions. Obviously in
Both results are in accordance with the classifi- multidimensional spaces there are many directions
cation of the SPO1 orbits presented in Sec. 3. Note along which one can depart from the SPO. In par-
that for both orbits of Fig. 11 the energy is con- ticular, in Fig. 12 we choose two such directions
served to very good accuracy as Er 10−8 up to described by two different types of ICs in the neigh-
t = 105 [see insets of Figs. 11(a) and 11(b)]. Based borhood of the SPO1 orbit. For the first type (IC1 )
on these results, as well as similar computations per- we perturb the positions of only the fixed parti-
formed for other N values (also for the Hollomon- cles by the same amount, attributing to each one of
type lattice, not presented here), we set Λ(1) = 10−4 these displacements a random sign, appropriately
as an empirical threshold value of the ftMLE for adjusting the position of the first particle in order
discriminating between regular (Λ(1) < 10−4 ) and to achieve the desired energy value. For the sec-
chaotic (Λ(1) ≥ 10−4 ) behaviors for orbits evolved ond approach (IC2 ), we perturb the momenta of all
up to t = 105 . Using this criterion, we were able particles, in the way described above, also correct-
to verify the validity of the power laws shown in ing the position of the first particle to achieve the
Figs. 7 and 9. appropriate energy.
In general, the majority of orbits in the vicinity In Fig. 12(a), we show the dependence of Dm
of a stable SPO are regular. Hence, the computa- on the system’s energy density for the SPO1 with
tion of their MLE allows us to estimate the “size” of N = 5 when the IC1 (blue curve) and the IC2 (green
regions of regular behavior around a stable periodic curve) are used. Both approaches produce values of
(a) (b)
Fig. 12. Dependence of the “size” Dm of the regular region around the stable SPO1 of the graphene-type Hamiltonian on
(a) the system’s energy density hN for N = 5 and (b) the number of degrees of freedom N for hN = 0.0039. In each panel
two types of ICs are considered: IC1 (blue curves) and IC2 (green curves) [see text for details]. The red dashed line in (a)
indicates the energy density hcN = 0.042 of the first SPO1 destabilization.
2030047-14
November 10, 2020 16:14 WSPC/S0218-1274 2030047
(a) (b)
Fig. 14. The spectrum of the averaged (over the final stage of their numerical evolution) LEs Λ(k) , k = 1, 2, . . . , N , of orbits
in the neighborhood of the unstable (a) SPO1 and (b) SPO2, for the graphene Hamiltonian with N = 17 and for various
energy levels E = E c17 + ΔE. E c17 = 0.18765 is the energy of the first destabilization of both SPOs. The dashed horizontal line
in both panels represents the level above which we consider a LE to be strongly positive. The error bars indicate one standard
deviation.
2030047-15
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
In order to investigate further the chaoticity of linear normal modes of the system. Furthermore,
orbits in the neighborhood of unstable SPOs we we wished to compare these two systems with what
consider the particular case of N = 17 for which is known in the literature for FPUT type lat-
the energy of first destabilization for both the SPO1 tices, whose (analytic) potentials consist of quartic
and the SPO2 orbits is E c17 ≈ 0.18765. Moving nearest neighbor interactions added to the harmonic
to higher energies, we explore the neighborhood ones.
of both unstable SPOs. In particular, we set the The two SPOs we chose to study are the
energy to be E = E c17 + ΔE (ΔE > 0) and con- ones that were also analyzed for FPUT systems
sider orbits starting in the immediate vicinity of in [Antonopoulos & Bountis, 2006; Antonopoulos
the SPOs. The initial conditions for these orbits et al., 2006; Bountis & Skokos, 2012]: The SPO1
are chosen so that their phase space distances from periodic solution, where every other particle is fixed
the SPOs are d ≈ 10−2 . This is achieved by start- while the ones about it perform the same oscil-
ing with the ICs of the SPO, perturbing the posi- lation x̂1 (t) in opposite directions, and the SPO2
tions of every fixed particle by the same small value solution, where between two stationary ones there
and appropriately changing the position of the first are two moving out of phase with respect to each
oscillator to retain the specific energy value. other, with the same x̂2 (t). These are continuations
The results of our numerical simulations are of the (N +1)/2 and 2(N +1)/3 linear normal modes
depicted in Fig. 14 where we plot the spectra of LEs respectively and are distinguished by the fact that
for various energy values for orbits in the neighbor- they are very easy to find: All one has to do is solve
hood of the unstable SPO1 [Fig. 14(a)] and SPO2 a single, second order nonlinear ODE for x̂1 (t) (or
[Fig. 14(b)]. In Fig. 14, we clearly see that close to x̂2 (t)). What is particularly interesting is that these
the destabilization energy, e.g. for ΔE = 0.0035 in two SPOs, although quite different from each other,
Fig. 14(a) (SPO1) and ΔE = 0.0061 in Fig. 14(b) share a lot of common dynamical properties.
(SPO2), small scale (“weak”) chaos occurs, charac- In the case of the graphene-type lattice they
terized by only the MLE being practically positive are stable at low energies and experience a first
(Λ(1) > 10−4 ). We note that for these two cases destabilization at energies per particle EN c /N that
the computed spectrum of LEs in not constantly decrease, as N increases, by power laws with nearly
decreasing, as Eq. (29) indicates. This is due to equal exponents, i.e. ∝ N −1.72 . This is quite dif-
well-known practical limitations of the two-particle ferent than the FPUT 1D lattices, for which this
method in accurately computing chaos indices, like decay is ∝ N −1 for the SPO1 solution and ∝ N −2
the LEs, for very weak chaotic behaviors [Mei & for the SPO2. On the other hand, the corresponding
Huang, 2018]. As the energy E increases, more LEs results for the Hollomon lattice are strikingly dis-
become larger than 10−4 indicating the onset of tinct: First of all, both SPO1 and SPO2 are unsta-
“strong” chaos in the neighborhood of both SPOs. ble at low energies and first stabilize along curves of
Thus, we conclude from these results that in this the form EN c /N ∝ N 2.68 ! This might be explained
case “strong” chaos is present for ΔE 0.004 in by the fact that the Hollomon term in the poten-
the case of SPO1 and ΔE 0.015 for the SPO2 tial has a power smaller than 2 and is dominant for
orbit. small energies, while for higher energies the har-
monic terms apparently become more important
and dominate the dynamics.
5. Discussion So much for local stability. Studying what hap-
In this work, we have studied two Hamiltonian sys- pens near the SPOs of the graphene model imme-
tems consisting of N particle systems in one dimen- diately after they become unstable, we have discov-
sion, whose interaction potential includes terms ered (just as in the case of FPUT systems) that
that are nonanalytic functions of the position coor- chaotic orbits do not immediately spread over large
dinates. The first one concerns “graphene-type” domains in phase space, but remain for long times
materials and the second MEMS satisfying Hol- close to the SPO exhibiting what one might call
lomon’s power-law of “work-hardening”. Our main “weak” chaos. For larger displacements, however,
purpose was to study their dynamics concentrat- (or longer integration times) nearby orbits eventu-
ing on the stability properties of two SPOs, which ally escape to much larger phase space domains of
are nonlinear continuations of the corresponding “strong” chaos.
2030047-16
November 10, 2020 16:14 WSPC/S0218-1274 2030047
2030047-17
November 10, 2020 16:14 WSPC/S0218-1274 2030047
A. Bountis et al.
Christodoulidi, H., Tsallis, C. & Bountis, T. [2014] Skokos, Ch. [2001] “On the stability of periodic orbits of
“Fermi–Pasta–Ulam model with long range interac- high dimensional autonomous Hamiltonian systems,”
tions: Dynamics and thermostatistics,” Europ. Phys. Physica D 159, 155–179.
Lett. 108, 40006. Skokos, Ch., Bountis, T. & Antonopoulos, C. [2007]
Christodoulidi, H., Bountis, T., Tsallis, C. & Drossos, “Geometrical properties of local dynamics in
L. [2016] “Dynamics and statistics of the Hamiltonian systems: The Generalized Alignment
Fermi–Pasta–Ulam β-model with different ranges of (GALI) method,” Physica D 231, 30–54.
particle interactions,” J. Statist. Mech. 12, 123206. Skokos, Ch. [2010] The Lyapunov Characteristic Expo-
Colombo, L. & Giordano, S. [2011] “Nonlinear elasticity nents and Their Computation, Lecture Notes in
in nanostructured materials,” Rep. Progr. Phys. 74, Physics, Vol. 790, pp. 63–135.
116501. Ustinov, A. V., Cirillo, M. & Malomed, B. A. [1993]
Esposito, P., Ghoussoub, N. & Guo, Y. [2010] “Math- “Dynamics in one-dimensional Josephson-junction
ematical analysis of partial differential equations arrays,” Phys. Rev. B 47, 8357(R).
modeling electrostatic MEMS,” AMS/Courant Instit. Wei, D., Kadyrov, S. & Kazbek, Z. [2017a] “Periodic
Math. Sci., New York. solutions of a graphene-based model in micro-electro-
Flach, S. & Gorbach, V. [2008] “Discrete breathers — mechanical pull-in device,” Appl. Comput. Mech. 11,
Advances in theory and applications,” Phys. Rep. 81–90.
467, 1–116. Wei, D. Skrzypacz, P. & Yu, X. [2017b] “Nonlinear waves
Hazim, H., Wei, D., Elgindi, M. & Soukiassian, Y. [2015] in rods and beams of power-law materials,” J. Appl.
“A lumped-parameter model for nonlinear waves in Math. 2017, 2095425.
graphene,” World J. Engin. Technol. 3, 57–69. Wei, D. & Liu, Y. [2012] “Analytic and finite element
Khan, Z. H., Kermany, A. R., Öchsner, A. & Iacopi, solutions of the power-law Euler–Bernoulli beams,”
F. [2017] “Mechanical and electromechanical prop- Finite Elements in Anal. Des. 52, 31–40.
erties of graphene and their potential application in Younis, M. I. [2013] MEMS Linear and Nonlinear Statics
MEMS,” J. Phys. D: Appl. Phys. 50, 053003. and Dynamics (Springer, Berlin).
Laskar, J. & Robutel, P. [2001] “High order symplec-
tic integrators for perturbed Hamiltonian systems,”
Celest. Mech. Dyn. Astron. 80, 39–62. Appendix A
Lee, C., Wei, X., Kysar, J. W. & Hone, J. [2008]
Linear Stability Analysis
“Measurement of the elastic properties and intrinsic
strength of monolayer graphene,” Science 321, 385. As is well-known, the standard approach to study
Lichtenberg, A. J. & Lieberman, M. A. [1986] Regular the stability of periodic orbits of 1D N -degree of
and Chaotic Dynamics (Springer, NY). freedom Hamiltonian lattices is through the method
Lu, Q. & Huang, R. [2009] “Nonlinear mechanics of of variational equations and monodromy matrix
single-atomic-layer graphene sheets,” Int. J. Appl. analysis of linear stability theory (see e.g. [Skokos,
Mech. 1, 443–467.
2001; Bountis & Skokos, 2012]). For completeness,
Macı́as Dı́az, J. E. [2017] “Numerical study of the pro-
we outline this approach in the present Appendix,
cess of nonlinear supratransmission in Riesz space-
fractional sine-Gordon equations,” Commun. Nonlin. as we applied it to the SPO1 and SPO2 solutions
Sci. Numer. Simulat. 46, 89–102. of the Hollomon lattice of Sec. 3.2 to compare with
Macı́as Dı́az, J. E. & Bountis, A. [2018] “On the trans- the predictions of our numerical criterion. Entirely
mission of energy in β-Fermi–Pasta–Ulam chains with analogous results were obtained for the SPO1 and
different ranges of particle interactions,” Commun. SPO2 of the graphene lattice. We were thus able
Nonlin. Sci. Numer. Simulat. 63, 307–321. to check that the analytical estimates regarding
Mei, L. & Huang, L. [2018] “Reliability of Lyapunov de(re)-stabilization energies, for both lattices, are
characteristic exponents computed by the two-particle very close to what one finds using the numerical
method,” Comput. Phys. Commun. 224, 108–118. criterion of Sec. 3. Thus, for most of the results
Oikonomou, Th., Nergis, A., Lazarides, N. & Tsironis, presented in this paper, we preferred to use the lat-
G. P. [2014] “Stochastic metastability by spontaneous
ter, as it is computationally much faster than linear
localisation,” Chaos Solit. Fract. 69, 228–232.
Pikovski, A. & Politi, A. [2016] Lyapunov Exponents. A
stability analysis.
Tool to Explore Complex Dynamics (Cambridge Uni- Let us recall first that to obtain numerically
versity Press). stable results when integrating Eq. (4), we need to
Shelupsky, D. [1959] “A generalization of the trigono- approximate the sign function by sgn(x − x0 ) ≈
metric functions,” The American Math. Month. 66, tanh[τ (x − x0 )] for a value of τ > 0 large enough
879884. (τ = 100 suffices). In that case, Eq. (4) takes the
2030047-18
November 10, 2020 16:14 WSPC/S0218-1274 2030047
∂Gj
(x = x̂) = k(δ ,j−1 − 2δ ,j +δ ,j+1 )
∂x
tanh[τ (x̂j − x̂j−1 )] τ
− λ |x̂j − x̂j−1 |q (δ ,j −δ ,j−1 ) q + 2
x̂j − x̂j−1 cosh [τ (x̂j − x̂j−1 )]
q tanh[τ (x̂j+1 − x̂j )] τ
− |x̂j+1 − x̂j | (δ ,j+1 − δ ,j ) q + , (A.3)
x̂j+1 − x̂j cosh2 [τ (x̂j+1 − x̂j )]
2030047-19