# Fermi-Pasta-Ulam nonlinear lattice oscillations

Thierry Dauxois and Stefano Ruffo (2008), Scholarpedia, 3(8):5538. | doi:10.4249/scholarpedia.5538 | revision #91257 [link to/cite this article] |

The **Fermi-Pasta-Ulam problem** bears the name of the three scientists who were looking for a theoretical physics problem suitable for an investigation with one of the very first computers, the *Maniac*. They decided to study the thermalization process of a solid. The Fermi-Pasta-Ulam (FPU) problem, first written up in a Los Alamos report in May 1955 (Fermi, Pasta, Ulam 1955), marked the beginning of both a new field, nonlinear physics (this problem is of central importance in the theories of solitons and chaos), and the age of computer simulations of scientific problems.
Mary Tsingou (now Mary T. Menzel), who took part in the numerical study, is not an author of the report, but her contribution is recognized by two lines of acknowledgment. It is time for a proper recognition of her contribution. Let us refer from now on to the **Fermi-Pasta-Ulam-Tsingou problem** (Dauxois 2008).

## Contents |

## The FPU paradox

The original idea, proposed by Enrico Fermi, was to simulate the one-dimensional analogue of atoms in a crystal: a long chain of particles linked by springs that obey Hooke’s law (a linear interaction), but with a weak nonlinear correction (quadratic for the FPU-\(\alpha\) model or cubic for the FPU-\(\beta\) model), see Figure 1.

A purely linear law for the springs guarantees that energy given to a single 'normal' mode always remains in that mode (see caption of Figure 2 for the definition of normal modes in terms of atom displacements from their equilibrium positions).

Fermi, Pasta and Ulam thought that, due to the nonlinear correction, the energy introduced into the lowest frequency mode \(k=\)1 should have slowly drifted to the other modes, until the equipartition of energy, a consequence of ergodicity, would have been reached. The beginning of the calculation indeed suggested that this was the case. Modes \(k=\)2, \(k=\)3,…, were successively excited, reaching a state close to equipartition, as shown in Figure 2. However, by accident, one day, they let the program run longer. When they realized their oversight and came back to the computer room, they noticed that the system, after remaining in the near equipartition state for a while, had then departed from it. To their great surprise, after 157 periods of the mode \(k=\)1, almost all the energy (all but 3%) was back to this mode. Further calculations, performed later with faster computers, showed that the same phenomenon repeats many times, and that a *super-recurrence* exists, at which the initial state is recovered with an even higher accuracy (see Ford, 1992 and Weissert, 1997 for a historical account).

Therefore, the system behaved in a surprising way: contrary to the predictions of statistical mechanics when the number of particles\(N\) is going to infinity, the energy equipartition state was not reached and energy was periodically returning to the initially excited \(k=\)1 mode. This highly remarkable result, known as the FPU paradox, shows that nonlinearity is not enough to guarantee the equipartition of energy.

## Solutions of the paradox

The solution of the paradox is two-fold. It requires the consideration of two important physical phenomena: the existence and stability of *solitons* in one dimension and the presence of *deterministic chaos*.

### In real space: generation of solitons

In 1965, using the so-called *continuum limit*, Zabusky and Kruskal (1965) were able to relate the periodic behavior observed in Figure 2 to the dynamics of localized excitations, nowadays known as *solitons*. Soliton generation was found for periodic boundary conditions, a situation which shares some features with the original fixed boundary conditions problem (Chirikov et al. 1973).
Starting from the FPU-\(\alpha\) equations of motion,
\(\tag{1}
\ddot u_n=(u_{n+1}+u_{n-1}-2u_{n})+\alpha \left[(u_{n+1}-u_{n})^2-(u_{n}-u_{n-1})^2\right]
\ ,\)

and restricting the investigation to long wavelength modes and right-propagating waves, they derived the Korteweg-de Vries (KdV) equation (Korteweg and De Vries 1895) \(\tag{2} w_\tau+\frac{1}{24}w_{\xi\xi\xi}+\alpha \,ww_\xi=0\ ,\)

where the field \(w\) is a conveniently rescaled spatial derivative of the displacement field \(u_n\ ,\) while \(\tau\) is a rescaled time and \(w_x=\partial w/\partial x\ .\) If the nonlinear force is cubic, i.e. for the FPU-\(\beta\) model, one ends up with a modified KdV equation (mKdV).

If the amplitude of the field \(w\) is large enough, the sinusoidal initial condition develops a sharp front (see Figure 3), which then breaks into a series of pulses, which are *solitons*. They preserve their shapes and velocities and, during their motion in the finite system with periodic boundary conditions, from time to time, they come back to the positions they had initially, restoring the initial condition. This is why FPU recurrence is observed. Quantitative estimates of the recurrence time were derived using soliton theory for the Toda lattice, which approximates well the FPU-\(\alpha\) model for small nonlinearities (Toda 1978).

It is important to emphasize that the FPU paradox would probably not have been a mystery for more than 10 years if, before Zabusky and Kruskal, somebody had got the idea to look carefully at the dynamics of the nonlinear lattice as a function of the space coordinate.

### In normal mode space: chaotic properties and Chirikov's resonance overlap criterion

Early attempts to explain the inefficient energy transfer among normal modes using linear resonance theory failed. A partial success was obtained using an adapted Krylov-Bogoliubov averaging method, which provided an estimate of the recurrence time (Jackson, 1963). Kolmogorov-Arnold-Moser (KAM) theorem had a strong impact on the debate around the FPU problem. KAM theorem states that most orbits of a slightly perturbed integrable Hamiltonian remain quasi-periodic. It looked natural to interpret the recurrent motion of normal modes in the FPU model as a consequence of the presence of such regular orbits. However, up to now, no quantitative estimate of the FPU recurrence time has been obtained using KAM theory. Moreover, the applicability of KAM theorem to the FPU Hamiltonian has been assessed only recently for the fixed end FPU-\(\beta\) model (Rink 2007), by proving that Birkhoff normal forms constitute an integrable approximation of the FPU Hamiltonian, the integrals being the linear energies of the normal modes. It has been also questioned whether the original FPU initial condition really lies or not on a KAM torus (Casetti et al 1997).

Compared with the rather inconclusive application of KAM theorem, a remarkable result was obtained by Izrailev and Chirikov (1966) applying the concept of *overlap of nonlinear resonances*. When the nonlinearity is small, one can separately consider each nonlinear resonance and apply perturbation theory. Besides KAM regular orbits, chaotic orbits with a positive Lyapunov exponent are present, but they are confined in thin layers and the Lyapunov exponent is small. On the contrary, for stronger nonlinearity, resonances are no more separated in frequency space and their overlap gives rise to strongly chaotic orbits, which have a larger Lyapunov exponent and diffuse in phase space. Using the resonance overlap criterion, Izrailev and Chirikov predicted that, as the nonlinearity of the FPU model is increased, by e.g. increasing energy, a transition should occur to a regime where relaxation to equipartition should be fast, as a consequence of the strongly chaotic nature of the orbits. This theoretical prediction was successfully tested numerically (Izrailev et al. 1970). Independendly, Bocchieri and collaborators (Bocchieri et al. 1970) showed the presence of the transition for a Lennard-Jones potential in one dimension. The *strong stochasticity threshold* (a term first used in Pettini and Landolfi 1990) exhibited in the numerical experiments of Livi and coworkers (Livi et al., 1985) is nothing but the transition from weak to strong chaos predicted by Chirikov. Nowadays we know that even below the strong stochasticity threshold, relaxation to equipartition occurs on longer and longer time scales as the energy is reduced, due to the overlap of higher order nonlinear resonances (De Luca et al. 1995, Shepelyansky, 1997).

The infinite-dimensional version of KAM-theory (KAM-tori are infinite-dimensional and stable) is found in the equation with FPU-nonlinearity by L. D. Pustyl'nikov in 1994 (See Reference below).

## Relaxation to equipartition for short wavelengths: chaotic breathers

If instead of preparing a long wavelength (low-frequency) initial state, one now puts the energy in the short wavelength (high-frequency) part of the normal mode spectrum (Zabusky and Deem 1967), the pathway to equipartition may lead to the creation of highly localized excitations that have an oscillating amplitude and bear most of the energy: they have been called *chaotic breathers* (Cretegny et al. 1998). Contrary to exact breathers (Flach and Willis 1998), which are periodic localised solutions whose existence has been proved also for FPU lattices (James 2001), these excitations display a chaotic evolution in space and time. This behavior is generic, because it is intimately related to *modulational instability*, a self-induced modulation of the steady state resulting from a balance between nonlinear and dispersive effects. This phenomenon appears in many branches of physics: fluid dynamics, nonlinear optics and plasma physics.

Figure 4 shows the time evolution of the \(\pi\)-mode, \(u_n(0)=a(-1)^n\ ,\) above the modulation instability critical amplitude \(a>a_c \approx \pi/(\sqrt{6}N)\ ,\) where \(N\) is the number of masses in a chain with periodic boundary conditions. Interestingly, this threshold amplitude coincides with the threshold estimate obtained using the Chirikov overlap criterion for short wavelength modes (Dauxois et al 2005a). The energy residing on site \(n\) is \(\tag{3} E_n = {1\over 2} \dot{u}_n^2 + {1\over 2}V(u_{n+1}-u_n) + {1\over 2} V(u_n-u_{n-1}), \)

where the FPU-\(\beta\) potential is \(V(x) = {1\over 2}x^2 + {1\over 4} x^4\ .\)

Figure 4 displays the full evolution, from the initial state to the generation of the moving chaotic breather. Figure 4(b), Figure 4(c) and Figure 4(d)
are three successive snapshots of the local energy \(E_n\) along the
chain. At short time, a slight modulation of the energy in the system appears (see Figure 4(b)) as a consequence of the destabilization of the \(\pi\)-mode. Later on, as shown in Figure 4(a), only a few breathers emerge. Inelastic collisions of breathers have a systematic tendency to favour the growth of the big breathers at the expense of small ones. Hence, in the course of time, the breather number decreases and only one, of very large amplitude, survives (see Figure 4(c)): this is the localized excitation called *chaotic breather*. The chaotic breather moves along the lattice with an almost ballistic motion: sometimes it stops or reflects back. During its motion the chaotic breather collects energy and its amplitude increases. It is important to note that the chaotic breather is never at rest and that it propagates with a given subsonic speed. Finally, the chaotic breather decays and the system reaches energy equipartition, as illustrated in Figure 4(d).

## Recent approaches

It has been observed that relaxation to equipartition for long wavelength initial conditions proceeds in two stages. On a short time scale a packet of low frequency normal modes is formed, with the higher modes cut-off exponentially (Fucito et al. 1982): this phase is associated with the formation of the soliton train of Figure 3. On longer times, energy is slowly flowing to high frequency modes. This scenario has been recently advocated by Galgani and coworkers (Berchialla et al 2004) on the basis of numerical simulations and of a resonant normal form approach to the KdV equation (Bambusi and Ponno 2006). Grist to this mill is also brought by the discovery of periodic orbits (so called *q-breathers*) that are exponentially localised in mode space (Flach 2005, Penati and Flach 2007).

An important open question is what remains of the FPU phenomenology in two-dimensional and three-dimensional lattices. Here results are scarce due to the numerical difficulty to simulate large lattices. Worth to mention is the work of Benettin (Benettin 2004), who argues against the two-stage relaxation picture.

## MATLAB Code

The following simple MATLAB code solves the equations for the dynamics of the FPU model and computes the normal modes (Dauxois et al. 2005b). It allows for the reproduction of different aspects of the FPU problem.

N=32; %Number of particles must be a power of 2 alpha=0.25; %Nonlinear parameter TMAX=10000; DT=20; tspan=[0:DT:TMAX]; options=odeset('Reltol',1e-4,'OutputFcn','odeplot','OutputSel',[1,2,N]); % Test different tolerances, changing Reltol for I=1:N, a=1; b(I)=a*sin(pi*I/(N+1)); b(I+N)=0; % FPU initial condition %a=1; b(I)=a*sin(pi*N*I/(N+1)); b(I+N)=0; % Zabusky-Deem init. cond. %k=0.8; sk=(sinh(k))\verb!^!2; ek=exp(-k); i1=I-N/4; i2=i1-N/2; %Solitons %b(I)=-0.5/alpha*log((1+exp(2*k*(i1-1)))/(1+exp(2*k*i1))); % Kink %b(I)=b(I)+0.5/alpha*log((1+exp(2*k*(i2-1)))/(1+exp(2*k*i2))); % Anti-kink %b(I+N)= sk*ek/alpha/cosh(k*i1)/(exp(-k*i1)+exp(k*i1)*exp(-2*k)); %b(I+N)=b(I+N)-sk*ek/alpha/cosh(k*i2)/(exp(-k*i2)+exp(k*i2)*exp(-2*k)); omegak2(I)=4*(sin(pi*I/2/N))\verb!^!2; % Mode Frequencies end [T,Y]=ode45('fpu1',tspan,b',options,N); % Time integration for IT=1:(TMAX/DT), TIME(IT)=IT*DT*sqrt(omegak2(1))/2/pi; % Time iteration loop YX(IT,1:N+1)=[0 Y(IT,1:N)]; YV(IT,1:N+1)=[0 Y(IT,N+1:2*N )]; sXF(IT,:)=imag(fft([YX(IT,1:N+1) 0 -YX(IT,N+1:-1:2)]))/sqrt(2*(N+1)); sVF(IT,:)=imag(fft([YV(IT,1:N+1) 0 -YV(IT,N+1:-1:2)]))/sqrt(2*(N+1)); Energ(IT,1:N)=(omegak2(1:N).*(sXF(IT,2:N+1).\verb!^!2)+sVF(IT,2:N+1).\verb!^!2)/2; for J=2:N-1, % Space loop DifY(IT,J)=Y(IT,J+1)-Y(IT,J); end end plot(TIME,Energ(:,1),TIME,Energ(:,2),TIME,Energ(:,3),TIME,Energ(:,4)); surf(DifY); % Space derivative field to show the soliton dynamics

fpu1 function function dy=fpu1(t,y); N=32;alpha=0.25; D(N+1)=y(2) -2*y(1)+alpha*((y(2)-y(1))\verb!^!2-y(1)\verb!^!2);D(1)=y(N+1); D(2*N)=y(N-1)-2*y(N)+alpha*(y(N)\verb!^!2-(y(N)-y(N-1))\verb!^!2);D(N)=y(2*N); for I=2:N-1, D(N+I)=y(I+1)+y(I-1)-2*y(I)+alpha*((y(I+1)-y(I))\verb!^!2-(y(I)-y(I-1))\verb!^!2); D(I)=y(N+I); end dy=D';

## References

Bambusi D. and Ponno A. (2006) *On metastability in FPU*, Comm. Math. Phys., 264:539-561.

Benettin G. (2004) *Time scales for energy equipartition in a two-dimensional FPU model*. Chaos, 15:015108.

Berchialla L., Galgani L. and Giorgilli A. (2004) *Localization of energy in FPU chains*. Discr. Cont. Dyn. Syst. A, 11:855-866.

Bocchieri P, Scotti A., Bearzi B and Loinger A. (1970) *Anharmonic chain with Lennard-Jones interaction*. Phys. Rev. A, 2:2013-2019.

Casetti L., Cerruti-Sola M., Pettini M. and Cohen E.G.D. (1997), *The Fermi-Pasta-Ulam problem revisited: Stochasticity thresholds in nonlinear Hamiltonian systems*,
Phys. Rev.E 55:6566.

Chirikov B., Izrailev F., and Tayurskij V. (1973) *Numerical experiments on the statistical behaviour of dynamical systems with a few degrees of freedom*,
Computer Physics Communications 5:11-17.

Cretegny T., Dauxois T., Ruffo S. and Torcini A. (1998) *Localization and equipartition of energy in the FPU-\(\beta\) chain: Chaotic breathers*. Physica D, 121:109-126.

Dauxois T. (2008) *Fermi, Pasta, Ulam and a mysterious lady*. Physics Today, 61, 1:55-57.

Dauxois T., Khomeriki R., Piazza F. and Ruffo S. (2005a), *The Anti-FPU problem*, Chaos 15:015110.

Dauxois T., Peyrard M. and Ruffo S. (2005b), *The Fermi-Pasta-Ulam numerical experiment: history and pedagogical perspectives*. European Journal of Physics 26:S3-S11.

De Luca J., Lichtenberg A.J., Lieberman M.A. (1995) *Time scale to ergodicity in the Fermi-Pasta-Ulam problem*. Chaos, 5:283-297.

Fermi E., Pasta J. and Ulam S. (1955) *Studies of nonlinear problems. I*. Los Alamos report LA-1940, published later in Collected Papers of Enrico Fermi, E. Segré (Ed.), University of Chicago Press (1965).

Flach S., Ivanchenko M.V. and Kanakov O.I. (2005) *q-breathers in Fermi-Pasta-Ulam chains: Existence, localisation and stability*. Phys. Rev. Lett., 95:064102-1-4.

Flach S and Willis C. R. (1998) *Discrete Breathers*. Phys. Rep., 295:181-264.

Ford J (1992) *The Fermi-Pasta-Ulam problem: Paradox turns discovery*. Phys. Rep., 213:271-310.

Fucito F., Marchesoni F., Marinari E., Parisi G., Peliti L., Ruffo S. and Vulpiani A. (1982) *Approach to equilibrium in a chain of nonlinear oscillators*, J. Physique (Paris), 43:707-713.

Izrailev F.M. and Chirikov B.V. (1966) *Statistical Properties of a Nonlinear String*. Soviet Phys. Dokl. 11:30.

Izrailev F.M., Khisamutdinov A.I. and Chirikov B.V., (1970) *Numerical experiments with a chain of coupled anharmonic oscillators*. LA-4440-TR, Los Alamos.

Jackson E.A. (1963) *Nonlinear coupled oscillators II. Comparison of theory with computer solutions*. J. Math. Phys., 4:686-700.

James G. (2001) *Existence of breathers on FPU lattices*. C.R. Acad. Sci. Paris Ser. I Math., 332:581-586.

Kolmogorov A.N. (1954) Dokl. Akad. Nauk SSSR, 98:527; Arnold V.I. (1963) Usp. Math. Nauk., 18:13-40; Moser J. (1962) Nachr. Akad. Wiss. Goettingen Math. Phys. Kl. 2, 1:15.

Korteweg D. J. and de Vries G. (1895) *On the Change of Form of Long Waves advancing in a Rectangular Canal and on a New Type of Long Stationary Waves*. Philosophical Magazine 5th Series, 36:422-443.

Livi R., Pettini M., Ruffo S., Sparpaglione M. and Vulpiani A. (1985) *Equipartition threshold in nonlinear large Hamiltonian systems: The Fermi-Pasta-Ulam model*. Phys. Rev. A 31:1039-1045.

Penati T. and Flach S. (2007), *Tail resonances of FPU q-breathers and their impact on the pathway to equipartition*. Chaos 17:023102.

Pettini M. and Landolfi M. (1990), *Relaxation porperties and ergodicity breaking in nonlinear Hamiltonian dynamics*, Phys Rev A 41:768.

Pustyl'nikov, L. D. (1997), *Infinitely dimensional nonlinear ordinary differential equations and KAM-theory*. Uspekhi Mat. Nauk, N 3(1997), pp. 105-160; english transl. in Russian Math. Surveys (1997), 52:3, 551-604.

Rink B. (2006) *Proof of Nishida's conjecture on anharmonic lattices*. Comm. Math. Phys., 261:613-627.

Shepelyansky D.L. (1997) *Low-energy chaos in the Fermi-Pasta-Ulam model*. Nonlinearity, 10:1331-1338.

Toda M. (1978) *Theory of nonlinear lattices*, Springer.

Zabusky N. J. and Deem G. S. (1967) *Dynamics of nonlinear lattices I, localized optical excitations, acoustic radiation and strong nonlinear behavior*. J. Comp. Phys. 2:126.

Zabusky N.J. and Kruskal M.D. (1965) *Interaction of "solitons" in a collisionless plasma and the recurrence of initial states*. Phys. Rev. Lett. 15:240-243.

Zabusky N.J., Sun Z., and Peng G. (2006) *Measures of chaos and equipartition in integrable and nonintegrable lattices*
Chaos 16:013130.

**Internal references**

- Jan A. Sanders (2006) Averaging. Scholarpedia, 1(11):1760.

- James Meiss (2007) Dynamical systems. Scholarpedia, 2(2):1629.

- Eugene M. Izhikevich (2007) Equilibrium. Scholarpedia, 2(10):2014.

- James Meiss (2007) Hamiltonian systems. Scholarpedia, 2(8):1943.

- Rob Schreiber (2007) MATLAB. Scholarpedia, 2(7):2929.

- James Murdock (2006) Normal forms. Scholarpedia, 1(10):1902.

- Jeff Moehlis, Kresimir Josic, Eric T. Shea-Brown (2006) Periodic orbit. Scholarpedia, 1(7):1358.

- Anatoly M. Samoilenko (2007) Quasiperiodic oscillations. Scholarpedia, 2(5):1783.

- Philip Holmes and Eric T. Shea-Brown (2006) Stability. Scholarpedia, 1(10):1838.

- David H. Terman and Eugene M. Izhikevich (2008) State space. Scholarpedia, 3(3):1924.

## Recommended Reading

Chaos (2005) Focus issue 15, devoted to the FPU problem, AIP Journal.

Dauxois T. and Peyrard M. (2006) *Physics of Solitons*, Cambridge University Press. ISBN: 0521854210

Gallavotti G. (Ed.) (2007) *The Fermi-Pasta-Ulam problem: A status report*, Springer-Verlag, ISBN: 3-540-72994-1.

Weissert T. P. (1997) *The Genesis of Simulation in Dynamics: Pursuing the Fermi-Pasta-Ulam Problem*, Springer.

## See Also

Discrete_Breathers,_Solitons,_and_Vortices_on_Anisotropic_Lattices, Korteweg-de Vries Equation and its Inverse Scattering Method, Solitons