The Rouse model[1] is the foundational basis for much of modern polymer physics. As seen in seminal early works[2,3], definitive research monographs[4], and textbooks, the Rouse model for an ideal polymer is said to describe polymer motion in lower-molecular weight polymer melts, to describe short-time and ’within the tube’ polymer motion of high-molecular-weight polymer melts, and, after inserting hydrodynamic interactions[5], to describe the motion of polymers in dilute solution.

Rouse[1] proposed a bead-and-spring model for an isolated polymer chain in solution, with individual beads subjected to uncorrelated hydrodynamic and random thermal forces, a harmonic force between pairs of bonded beads, and sufficient hydrodynamic drag that bead inertia could neglected. Rouse’s model leads to a set of coupled linear differential equations for the motions of the beads. The solutions to these differential equations are the Rouse modes, whose mean-square amplitudes are predicted by the model and whose amplitudes on the average decay exponentially with time. From the mode behavior and various mechanical assumptions, the Rouse model predicts the polymer’s diffusion coefficient, the polymer’s contribution to the solution viscosity, and the motion of the polymer’s end-to-end vector.

There are other models for polymer dynamics and polymers in solution. In particular, Kirkwood and Riseman[6] proposed a bead-and-spring polymer model whose description of the important polymer motions is entirely opposite to that of Rouse. This author[7,8] has extended the Kirkwood-Riseman model to concentrated polymer solutions.

One path to testing these models is the use of computer simulations. Even limiting our review to molecular dynamics methods, one readily identifies well over a thousand articles that might be included in a comprehensive review of simulations of polymer dynamics, not counting research approaching engineering applications of simulations. Simulations using Monte Carlo methods, in which momentum variables are suppressed, are not relevant to tests of dynamics. In addition, also not considered here, there is an extremely large but non-communicating research field on computer simulations of the dynamics of biopolymers and structures, e.g., complete ribosomes.[9]. Finally, going back two-thirds of a century, applications of computational methods to a Rouse-like model (albeit with a much more elaborate system Hamiltonian) of simple organic molecules were used to interpret their infrared and Raman vibrational frequencies.

There is disagreement in the literature as to the validity of the Rouse model and its utility for describing polymer dynamics. The original simulation of a single Rouse chain by Grest and Kremer[10] did confirm the model. For melts, matters are more complicated. Tsolou, et al.,[11] say of their simulations “...The Rouse Theory is found to provide a satisfactory description of the simulation findings, especially for rings with chain length between C${}_{50}$ and C${}_{170}$...”. Roh, et al.,[12] conclude that “…the structure and relaxation of the unentangled short-chain-branched ring and linear melt systems can be reasonably well characterized with the Rouse model, regardless of the short branches…". Tsalikis, et al.,[13] conclude “...Overall, our MD simulations have demonstrated the validity of the Rouse model for the dynamics of the simulated PEO ring melts...” Kopf, et al.,[14] report that the Rouse model remained accurate in blends of light and heavy polymer chains.

Some authors describe the Rouse and reptation models as assumptions. For example, Colmenero[15] observes “Nowadays it is generally assumed that the Rouse model provides a suitable description of chain dynamics of unentangled polymer melts...". Colmenero continues by noting that the reptation model assumes the validity of the Rouse model at short times. He then notes the Rouse model’s failures, and the model predictions that the failures invalidate.

On the other hand, in his review Viscoelasticity and Molecular Rheology in Polymer Science: A Comprehensive Reference[16], Likhtman observes “We note that often models are studied by theoreticians just because they are analytically solvable and used by experimentalists because of availability of analytic solutions...”, leading to his conclusion “This coupling [GP: between Rouse modes; see below] suggests that the Rouse mode description is not very useful for entangled polymers.” Kalathi, et al.,[17] note Likhtman’s observation, yet use a Rouse mode analysis in their work, saying as a sensible defense that experimentalists “...still tend to model chain dynamics in the language of the Rouse model. Understanding experimental results therefore requires us to analyze the simulations in the same manner."

There are two fundamental sorts of tests of any model of polymer dynamics, namely direct tests and inferential tests. In direct tests, one examines what the model actually says about polymer motions. The Rouse modes either do or do not have the behavior predicted by the model. In inferential tests, one compares the model’s predictions for measurable parameters with the experimental behavior of the parameters. A well-known inferential test of the Rouse model is the dependence of the melt viscosity on the polymer molecular weight M. The observed molecular-weight-dependence for melts of lower-molecular-weight polymers agrees with calculations based on the model, leading one to observe that the model agrees with experiment. The power of inferential tests can be overstated. The observed M-dependence would only be a demonstration that polymer dynamics are Rouse-like if the prediction could be shown to be unique, i.e., the molecular weight dependence would only show that Rouse dynamics are correct if one could show that no fundamentally different model makes the same prediction. As it happens, there is actually a fundamentally different model of polymer dynamics, the Kirkwood-Riseman model with hydrodynamics suppressed, that predicts the same M-dependence, so the inferential test is non-informative.

This review focuses on simulations that make direct tests of the Rouse modes, their amplitudes, and the time dependence of their relaxations.We identified a substantial number of these simulations. To the limited extent that we found simulational results that did not agree with our conclusions, we were careful to include those results.

In the following, Section 2 of the paper describes the Rouse and Kirkwood-Riseman models. These models are familiar from the literature, so we limit ourselves to descriptions of these models that are adequate to establish notation and terminology. Section 3 presents some general tests for the models. The following Section 4 describes seriatim the simulation studies being reviewed and what they reveal. Almost all of these works were concerned with polymer melts. The points at which the Kirkwood-Riseman model differs from the Rouse model correspond to questions that have rarely been asked in simulations; one study (which happens to support the Kirkwood-Riseman model) is noted. An extended Discussion unifying these results follows as a further section, Section 5. Some readers may prefer to skip to Section 5 and then return to read fine details in Section 4. The Conclusions in Section 6 close the paper.

Here we treat the Rouse and Kirkwood-Riseman models. This review is primarily about the widely-used Rouse model, not the relatively rarely encountered Kirkwood-Riseman model, which has had few simulational tests.

The Rouse and Kirkwood-Riseman models for polymer dynamics both describe a polymer chain as a line of N hydrodynamically active beads, labelled $\{0,1,2,\dots ,N-1\}$ in order along the chain, linked by hydrodynamically inert Hookean springs. The beads have cartesian coordinates $({\mathbf{r}}_0,{\mathbf{r}}_1,\dots {\mathbf{r}}_{N-1})$ and hydrodynamic drag coefficients $(f_0,f_1,\dots ,f_{N-1})$. The springs, with force constants $k_i$, serve to control the average distance between bonded pairs of beads. The hydrodynamic force on a bead i is taken to be linear in the velocity of the bead with respect to the solvent, namely Here ${\dot{\mathbf{r}}}_i$ is the velocity of bead i, while ${\mathbf{v}}_i$ is the velocity that the fluid would have had, at the location of bead i, if the bead were absent.

In most treatments, the bead drag coefficients $f_i$ and force constants $k_i$ are taken to have common values f and k, respectively. f is taken to be large enough that bead motions are overdamped on time scales of interest, so that bead inertia can be neglected. In a few studies, simulations in which some beads have large sizes have given physically interesting results. When the solvent exerts a force on a bead, from Newton’s Third Law that bead exerts an equal and opposite force on the solvent. In the Zimm and Kirkwood-Riseman models, but not the Rouse model, the forces that the beads exert on the solvent create solvent flows that perturb the motion of the solvent around each of the other beads. These perturbations are the bead-bead hydrodynamic interactions, described in both the Zimm and the Kirkwood-Riseman models by the Oseen tensor.

The polymer coils are part of a thermal system. Corresponding to the frictional forces of equation 1, the fluctuation-dissipation theorem guarantees that there must on each bead be a fluctuating thermal force ${\mathbf{F}}_i\left(t\right)$, the thermal forces serving to maintain the temperature of the system. In the Rouse model, there are no hydrodynamic interactions between the beads, so the ${\mathbf{F}}_i\left(t\right)$ on different beads are uncorrelated. In the presence of hydrodynamic interactions, the fluctuating thermal forces on different beads must necessarily have cross-correlations. These considerations lead to equations of motion for the beads, in the form of a set of coupled linear differential equations with constant coefficients. The end beads of linear chains are special cases. For the other beads, one has

Solution of the Rouse model begins with the Rouse modes, which are a discrete Fourier transform between an index i that labels the position coordinates ${\mathbf{r}}_i\left(t\right)$ of the beads and an index p that labels the N Rouse modes ${\mathbf{X}}_p\left(t\right)$. The Rouse modes, like the position coordinates, form a complete orthogonal set of coordinates that between them specify the positions of the beads in a polymer chain. For a linear chain, ${\mathbf{X}}_p$ is related to ${\mathbf{r}}_i\left(t\right)$ by with an inverse The Cartesian representation of the $p^{th}$ normal mode is found from equation 4 by setting one Rouse coordinate ${\mathbf{X}}_p\left(t\right)=1$ and setting all the other Rouse coordinates to zero.

The ${\mathbf{X}}_p\left(t\right)$ are the normal modes of the Rouse model. In the Rouse model:

In particular, and where b is the root-mean-square average distance between two neighboring beads, $k_B$ is Boltzmann’s constant, T is the absolute temperature, and the two approximations are valid for small $p/N$. The $X_p\left(0\right)$ are predicted to have independent Gaussian random distributions, so all higher moments of $X_p\left(0\right)$ can be calculated from $\langle {\left(X_p\left(0\right)\right)}^2\rangle$. The statement that the modes are orthogonal at all times, so that $\langle X_p\left(0\right)X_q\left(t\right)\rangle =0$ if $p\ne q$ arises from the forces in the Rouse model, and is not equivalent to the equally-correct statement that the Rouse coordinates are orthogonal. It is straightforward to make a modest modification of the Rouse model such that some modes become cross-correlated, namely one applies to the molecule a time-independent external shear field.[18]

From the Rouse modes, one may obtain predictions for the self-diffusion coefficient, the viscoelastic behavior, mean-square bead displacements, and the relaxation of the polymer end-to-end vector. These predictions are largely outside the scope of this review.

Many analyses begin by assuming the fundamental validity of the Rouse[1] and Zimm[5] models of polymer dynamics, at least on some time and distance scales. It is not always recalled that the Rouse and Zimm models were preceded by the Kirkwood-Riseman model[6]. When the Kirkwood-Riseman model is mentioned at all, it tends to be treated as being much the same as the Rouse and Zimm models. As a grain of truth, these three models all describe a single, isolated, polymer coil, not a polymer melt.

Kirkwood and Risemann discuss the motions of a polymer coil in a shear flow. The polymer has a center of mass ${\mathbf{R}}_{\mathrm{CM}}\left(t\right)$. In the Kirkwood-Risemann model, the velocity of a bead i is Here $\frac{d{\mathbf{R}}_{\mathrm{CM}}\left(t\right)}{dt}$ is the average translational velocity of all the beads, $\mathsf{\Omega }$ is the polymer’s rotational velocity, and ${\mathbf{u}}_i\left(t\right)$ is the part of the bead’s velocity due to chain internal modes. In the remainder of Kirkwood and Riseman’s calculations, chain internal modes and the velocities they create are ignored, as providing only secondary corrections. $\frac{d{\mathbf{R}}_{\mathrm{CM}}\left(t\right)}{dt}$ and $\mathsf{\Omega }$ are determined by the restriction that the system is heavily overdamped, meaning that the net force and the net torque on the polymer must, except on very short time scales, average very nearly to zero. In a quiescent fluid, one then necessarily has $\frac{d{\mathbf{R}}_{\mathrm{CM}}\left(t\right)}{dt}=\mathbf{0}$ and $\mathsf{\Omega }=\mathbf{0}$. In a simple shear field $d{v}_x/dy=\mathrm{constant}$, it is in general impossible for every bead of a polymer chain to move at the velocity given by the local value of the shear field. As a result, the polymer translates at the average of the fluid speeds seen by all the beads. The polymer also rotates, with torques due to the x and y components of the rotational velocity of all the beads on the average canceling. However, at any time most beads are moving with respect to the neighboring fluid, the dissipation arising from the corresponding viscous drag being the viscosity increment created by the polymer. In the absence of a shear flow, a Kirkwood-Riseman model chain performs translational and rotational diffusion, the latter having the effect of relaxing the direction of the chain end-to-end vector.

It is generally ignored that the Kirkwood-Riseman and Rouse models give completely contradictory descriptions of how polymer coils move in solution and create viscous dissipation. Consider a linear chain having N beads. In all models, the polymer coil has three center-of-mass coordinates with a center-of-mass velocity, corresponding to an average translational motion of the entire chain. In the Rouse model, the Rouse transformation replaces the $3N$ Cartesian coordinates with three center-of-mass coordinates and $3N-3$ internal Rouse coordinates $X_{p\alpha }\left(t\right)$. Here $p\in (1,N-1)$ and $\alpha \in (x,y,z)$. In the Rouse model, each $X_{p\alpha }\left(t\right)$ corresponds to a Rouse mode. The $3N-3$ Rouse modes have a common feature, namely that in each mode at least some of the beads move with respect to each other.

In contrast to the Rouse and Zimm models, in the Kirkwood-Riseman model an N-bead polymer coil has three translational modes, describing an averaged translation of the polymer chain, and three rotational modes, describing an averaged rotation of the polymer chain. In the translational and rotational components of the bead motions, the distances between the polymer beads remain constant. There then remain $3N-6$ internal modes in which the relative positions of the beads change with time. The Rouse and Kirkwood-Riseman models thus do not agree as to how many internal modes, modes in which the beads move with respect to each other, a polymer coil has. One model says $3N-3$ modes, while the other says $3N-6$ modes.

It should have been, but was not, immediately apparent that the Rouse model with its $3N-3$ independent internal modes are inconsistent with basic classical mechanics, in which an N-atom molecule can translate and rotate, and therefore has $3N-6$ internal degrees of freedom. This count of the allowed number of internal modes is established with absolute certainty by experimental and theoretical studies of infrared and Raman spectroscopy. Furthermore, prominently from Raman spectroscopy of molecular crystals, rotational modes are slow relative to most vibrational modes, so it would be incorrect to propose that the discrepancy in the mode count can be hidden in a few high-frequency modes that are elsewise of no significance.

The Rouse and Kirkwood-Riseman models are also entirely opposite in their descriptions of how polymeric viscosity increments arise. Rouse assigns viscous dissipation to the polymer chain’s internal modes, which lead to forces opposing the hydrodynamic forces due to the shear field, while denying whole-body rotation in a shear field. In a shear field, in Rouse’s model a polymer is subject to an affine deformation. Its internal forces due to the deformation lead to viscous dissipation. Kirkwood and Riseman assign viscous dissipation to whole-body rotation, while neglecting internal motions of a chain as providing only secondary corrections.

In the Rouse model, the correlation functions $\langle X_{p\alpha }\left(t\right)X_{p\alpha }\left(0\right)\rangle$ relax exponentially in time. Contrariwise, if mode relaxations are stretched exponentials in time, or have some other time dependence, then assuredly the underlying polymer dynamics are not those of the Rouse model. In each Rouse mode, most beads are displaced from their rest positions and return back to them as time advances. In the Rouse model, all beads have the same rest position; at rest, all beads are at the center of mass. Rouse modes correspond to spatially (but not temporally) oscillatory displacements having larger or smaller wavelengths. Rouse modes are not wavelets; they do not refer to fluctuations in a single localized region of a chain. The Zimm model[5] is substantially similar to the Rouse model, except that bead-bead hydrodynamic interactions at the level of the Oseen tensor are included in the calculation. These hydrodynamic interactions change how rapidly each mode relaxes, but the internal modes of the Zimm model are the same as the internal modes of the Rouse model except for their relaxation rates.

Rouse applied his model[1] to calculate the viscosity increment created by a Rouse-model polymer. An applied shear field, with fluid velocity along the x axis, velocity gradient along the y-axis, and vorticity vector in the z direction, was claimed to displace polymer beads, but according to Rouse only in the x direction, the direction of the velocity. Motions of the polymer beads in the y and z directions were claimed not to be affected by this shear field. In the Rouse model, a polymer chain subject to an external shear field thus performs an affine deformation, as described more recently by deGennes[2], but in the Rouse model polymer chains in a shear field do not rotate. The forces of the polymer beads on the solvent, due to their having been displaced relative to each other by the shear, as described by Rouse modes, lead in the model to viscous dissipation.

In the Rouse model every monomer bead is exposed to a Gaussian random force. The displacements ${\mathbf{r}}_i\left(t\right)-{\mathbf{r}}_j\left(0\right)$ arise from weighted linear sums of these external forces, so ${\mathbf{r}}_i\left(t\right)-{\mathbf{r}}_j\left(0\right)$ must in the Rouse model have a Gaussian random distribution, leading to and therefore This is the Gaussian approximation for the intermediate structure factor $g^{\left(1\right)}(q,t)$; its time dependence is determined by the one- and two-particle time-dependent mean-square particle displacements. [Some authors denote $g^{\left(1\right)}(q,t)$ as $S(q,t)$.] Chong and Fuchs[19] offer a demonstration that this Gaussian approximation is theoretically appropriate. We return later to the accuracy of this approximation as tested by simulations.

An interesting physical question is the possible presence of cross-correlations in Rouse mode amplitudes in physical systems, i.e., is $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$ ever non-zero for $p\ne q$ and/or $i\ne j$? Such correlations do not exist in a Rouse-model polymer but could exist in some system that is not described by the Rouse model. A plausible general form for a cross-time-correlation function of two Rouse modes is A corresponding equation, not the most general one, for cross-correlations between the mean-square amplitudes of two Rouse modes would be The elaborate normalization seen here is invoked because even a variable whose typical size is small can still make a strong contribution, relative to its size, to the variable with which it is correlated. From the above definition, ${\mathsf{\Phi }}_{ppii}\left(0\right)=1$; a variable is perfectly correlated with itself. In an equilibrium, nonchiral fluid, from reflection symmetry ${\mathsf{\Phi }}_{pqij}\left(0\right)=0$ if $i\ne j$. For an isolated Rouse chain in a fluid with shear, this author has previously shown from simulations that ${\mathsf{\Phi }}_{ppij}\left(t\right)\ne 0$ can occur. In the presence of shear, one finds cross-correlations for $p=q$ and $i\ne j$.[18]

In many cases, $\langle X_{pi}\left(t\right)X_{pi}\left(0\right)\rangle$ is found simulationally to decay as a stretched exponential in time, i.e., Some authors have proposed that a characteristic time ${\tau }_p$ may be extracted from this form, by analogy with the simpler integral namely or where here $\mathsf{\Gamma }$ is the gamma function. No source appears to have proposed a physical basis for identifying ${\tau }_p$ or $\overline{\alpha }$ rather than some other average over $\langle X_{pi}\left(t\right)X_{pi}\left(0\right)\rangle$ as an appropriate characteristic time, but this average is simple and broadly used.

The effective relaxation rate for this stretched exponential in time is[20] For the Rouse model, $W^{\mathrm{eff}}$ only depends on the temperature, the monomer friction factor f, and the segment length b.

This Section considers a few general tests of the Rouse model.

First, Rouse assumed that a shear velocity in the x direction would only induce bead displacements parallel to the x axis. Furthermore, under the influence of a shear field, in the Rouse model the modes remain uncorrelated, while the relaxation rates are independent of the applied shear. Rouse’s formula for the polymeric viscosity increment is based on these assumptions. In contrast, in the Kirkwood-Riseman model a shear field causes polymers to rotate. An applied shear field $d{v}_x/dy$ creates bead motions parallel and antiparallel to the x axis and also parallel and antiparallel to the y axis. Kirkwood and Riseman’s model does not consider the response of polymer internal modes to an applied shear.

To resolve this contradiction between the Rouse and Kirkwood-Riseman models, I made Brownian dynamics simulations on a single bead-spring polymer coil[18]. Hydrodynamic interactions were not included, so the random thermal forces on separate beads could be treated as being uncorrelated. As the test was of the Rouse model, the original Rouse harmonic potential linked all pairs of bonded beads. Chains had no bending constraints. Remote parts of the chain were able to pass ghostlike through each other.

It was shown that the Kirkwood-Riseman model, so far as it goes, is correct, while the Rouse and Zimm models are wrong for a polymer coil in a shear field. In particular, these simulations demonstrated for coils in a shear field: Polymer coils do indeed rotate, so that where $x_i$ and $y_i$ are the x and y components of bead i’s location relative to the polymer center of mass, while $v_{xi}$ and $v_{yi}$ are the x and y components of bead i’s velocity. In addition, in a shear field the Rouse modes become cross-correlated, so the Rouse coordinates cease to represent normal modes. Under shear, the mean-square amplitudes and relaxation rates of the Rouse modes are found to depend on the shear rate. All these results are inconsistent with the Rouse model.

The deficiency in the Rouse model is at its very beginning. Its equations of motion for the polymer beads have no applied shear, so the model only refers to an isolated polymer coil in a quiescent liquid. Rouse polymers therefore do not rotate. However, they are also not subject to a shear field, so they do not create viscous dissipation.

Arising from these theoretical models is the question of what one means by a ’bead’. Conventional polymer molecules are not actually formed from little spheres connected by very thin Hooke’s-law springs. The beads and springs are abstracted from an actual description of a polymer molecule. The Rouse model is not an atomic-level model; rather, atomic motions involving distances shorter than some limiting length have been approximated into the beads and springs. It has conventionally been assumed that the limiting length is the Kuhn length, but that assumption is not intrinsic to the Rouse model. In the Kuhn model, a polymer is divided into N segments of length b, the segments being straight and `freely-jointed’, i.e., each segment is free to make an arbitrary angle with the next segment in line. N and b are determined by two constraints, namely that the length of the fully-stretched polymer is $Nb$, while the root-mean-square polymer end-to-end distance R satisfies $\langle R^2\rangle =N{b}^2$. The statement that the length is $Nb$ implies that, within each segment, the polymer is fully stretched. On the other hand, in the Kirkwood-Riseman model, the beads are defined to be monomers, the long-range nature of the Oseen hydrodynamic interaction tensor leading to the result that the distribution function for the distances between relatively remote beads is the significant distribution function. In a simulation, no approximations are required; one can actually simulate a Rouse-model chain composed of hydrodynamic beads connected by Hookean springs.

Agapov and Sokolov[21] note that these definitions of N and b refer purely to polymer statics, but that a dynamic bead of the Rouse model has often been identified with a Kuhn segment. Agapov and Sokolov compare various implicit determinations of bead size with the nominal Kuhn length b. The notion of the identification was that the Kuhn length was the length of the shortest chain segment that followed Rouse dynamics, at least in the melt, and that the dynamics of shorter chain segments were not described by the Rouse model. As Agapov and Sokolov explain, for a Rouse chain the relaxation rate of the intermediate structure factor $g^{\left(1\right)}(q,t)$, typically as obtained from neutron scattering, scales as $q^4$. However, if the scattering vector q is made sufficiently large, one is probing chain motions over distances less than that for which the Rouse model is valid, in which case at some $q=Q$ the relaxation rate deviates noticeably from $q^4$ behavior. The corresponding distance $2\pi /Q$ defines a dynamic bead size. Agapov and Sokolov note the analysis of Nicholson, et al.,[22] that in poly(dimethylsiloxane) $2\pi /Q\approx b\approx 16\text{Å}$ but that in polystyrene $2\pi /Q\approx 55\text{Å}$ is about 2.5 times as large as b, i.e., the dynamic bead size inferred from $g^{\left(1\right)}(q,t)$ is perhaps 2.5 times the Kuhn length. Agapov and Sokolov also note computer simulations[23] and oscillatory flow birefringence studies[24] that found a dynamic bead size that is considerably larger than the Kuhn length. They conclude that interpretations of the Rouse model that require that the dynamic bead must be the size of a Kuhn length cannot be correct.

Colmenero[15] remarks that the reptation model assumes the validity of the Rouse model at times shorter than the entanglement time ${\tau }_e$. He observes that the two fundamental results of the Rouse model are that the Rouse modes are independent, so that $\langle X_{pi}\left(0\right)X_{qj}\left(0\right)\rangle =0$ if $p\ne q$, and that the Rouse correlators $\langle X_{pi}\left(0\right)X_{pi}\left(t\right)\rangle$ decay exponentially in time. Colmenero notes a variety of cases in which the observed correlators relax as stretched exponentials rather than exponentials in time, for example in cold melts[25] and in blends[26,27,28]. As an interpretation, he suggests that, at low temperatures, non-exponential decay of Rouse correlators might arise from coupling of polymer motions to local density fluctuations (the $\alpha$ relaxation). Issues then arise from the non-exponential time dependence of the Rouse correlation functions $\langle X_{pi}\left(0\right)X_{pi}\left(t\right)\rangle$. Colmenero emphasizes that various time-dependent physical quantities, such as coherent and incoherent scattering functions and dielectric relaxation spectra, have historically within the Rouse model been calculated incorrectly, because the calculations incorrectly assume exponential relaxation of the $\langle X_{pi}\left(0\right)X_{pi}\left(t\right)\rangle$. Therefore, inferential tests of the Rouse model, when based on comparisons of experimental data with Rouse’s formulae for the zero-shear viscosity and other experimental quantities, are invalid because the Rouse correlators do not decay exponentially.

Colmenero[15] proposed to interpret the non-exponential dependence of $\langle X_{pi}\left(0\right)X_{pi}\left(t\right)\rangle$ by replacing the Rouse model’s Langevin equation of motion with a Generalized Langevin Equation. Generalized Langevin Equations are a natural outcome of the Mori formalism[29,30]; applications to polymer dynamics have been explored by Guenza and colleagues[31,32,33]. Colmenero writes for the time evolution of a Rouse amplitude correlator in which $\mathsf{\Gamma }(t-s)$ is a memory function. (An integration by parts would in the integral replace the time derivative of $\langle X_{pi}\left(0\right)X_{pi}\left(s\right)\rangle$ with the function itself.) To solve this equation, $\mathsf{\Gamma }(t-s)$ was assumed to be short-lived, so that the convolution integral became nearly a single-time product, so on defining $\xi \left(t\right)={\int }_0^{t}ds\mathsf{\Gamma }\left(s\right)$, an approximate solution was proposed to be On requiring this equation to yield a stretched exponential in t, Colmenero proposed that, in the time regime in which $\xi \left(t\right)\gg f$, the simplest form for $\xi \left(s\right)$ is a power law in s. He uses his results to derive a stretched-exponential form for the time autocorrelation function $\langle {\mathbf{R}}_E\left(0\right)·{\mathbf{R}}_E\left(t\right)\rangle$ of the polymer end-to-end vector ${\mathbf{R}}_E$. Comparison was then made with prior atomistic molecular dynamics simulations[27,34] of polyethylene oxide and polymethylmethacrylate/polyethylene oxide melts. $\langle {\mathbf{R}}_E\left(0\right)·{\mathbf{R}}_E\left(t\right)\rangle$ and the self part of the dynamic structure factor at a series of temperatures were successfully fit to the predicted stretched-exponential time dependences. Values of ${\tau }_p$ and $\beta$ at each temperature, as obtained independently from the two physical quantities, were said to be in rather good agreement. Colmenero also compared his results with the Ngai coupling model.[35,36].

A rarely-tested prediction of the Rouse model states that Rouse modes are orthogonal in the sense that $\langle X_p\left(t\right)X_q\left(0\right)\rangle =0$ if $p\ne q$. There have been several successful tests of this prediction for the special case $t=0$, including results of Kopf, et al.,[14] and Tsalikis, et al.[13]. However, as part of an extended review of theoretical models for viscoelasticity and molecular rheology, Likhtman[16] obtained $\langle X_1\left(t\right)X_3\left(0\right)\rangle$ for several models of a melt of entangled polymers. While $\langle X_1\left(0\right)X_3\left(0\right)\rangle \approx 0$ to good approximation, with increasing t Likhtman found (his Figure 33) that $\langle X_1\left(t\right)X_3\left(0\right)\rangle$ increases substantially with increasing t, to far above any noise in the simulation, and then fades away. Likhtman’s result is entirely contrary to expectations from Rouse model dynamics, in which modes are not cross-correlated at any time, leading Likhtman to his observation as quoted above that “This coupling suggests that the Rouse mode description is not very useful for entangled polymers."

In this Section, we have considered the response of a polymer chain to shear, the mapping of Rouse beads onto polymer segments, implications of non-exponential decays of Rouse correlators, and tests of the orthogonality of Rouse modes. None of the outcomes entirely supported the Rouse model or its interpretations.

We now turn to simulations that actually examined properties of Rouse modes.

Dynamic simulations of polymers are readily traced back to the work of Grest and Kremer[10], who simulated a bead-spring model for a polymer chain, in which the beads are subject to independently fluctuating thermal forces, all bead pairs separated by less than a specified distance interact with a Lennard-Jones potential, and each bead’s motions are coupled to a heat bath that supplied a friction term and a thermal driving force. The interaction between bonded beads was represented with a finitely extensible (FENE) potential where $r_{ij}$ is the distance between two beads, k and $R_o$ are simulational parameters, and the potential is set to zero for $r_{ij}>R_o$. The above potential is not the harmonic potential used by Rouse, so strictly speaking this simulation was not a test of the Rouse model. Single chains and rings containing 50-200 beads, and a 200-bead chain with no Lennard-Jones potentials, were examined. The diffusion coefficient, bead mean-square displacement, mean-square center-of-mass displacement, and static structure factor $S\left(q\right)$ were calculated. Comparisons were made with theoretical expectations. Chain motions for these single chains were found to be consistent with the Rouse model.

Kremer and Grest[37] then reported their pioneering study of a melt of bead-spring polymers, covering from short nonentangled (`Rouse’) regime up to the highly entangled (`reptation’) regime. In their molecular dynamics simulation, all beads exerted a purely repulsive Lennard-Jones potential and had a FENE bond between next neighbors along each polymer chain, and had a weak frictional force $-f\mathbf{v}$ and a corresponding thermal force. Systems with chain lengths N from 5 to 400 beads and a total of 250 to 20,000 beads, corresponding to 16 to 100 chains, were studied. The nominal entanglement length was reported to be $\approx 35$ beads, so these simulations included both unentangled and entangled systems. Static properties that were examined include the mean-square end-to-end distance, the radius of gyration, the static structure factor, and the mean-square amplitude of Rouse modes. These quantities showed the expected scaling dependences on N. Time-dependences of the mean-square displacements of single monomers, chain centers-of-mass, and monomers relative to the center of mass, of Rouse amplitudes, of scattering functions, and of chain motion relative to a primitive path were also analyzed. Here we focus on the Rouse modes.

Kremer and Grest reported the normalized Rouse mode temporal autocorrelation functions for chain lengths $N=20,50,100,$ and 200. Figure 1 shows their results. Kremer and Grest interpreted these as clear single exponentials for the two shorter chains and two time scales for the longer chains. The figures show our fits to stretched exponentials in time. The relaxations are fit well by stretched exponentials. In these pioneering studies, for the two longer chains, at long times the correlation functions show weak fluctuations on top of the stretched exponentials, making it difficult to determine $\beta$ accurately. Kremer and Grest also evaluated $\langle X_p\left(0\right)X_q\left(0\right)\rangle$ for $p\ne q$, finding that the static cross-correlations vanish to within the noise in the simulations.

Tsalikis, et al.[13] report simulations of ring polymers. Their study is noteworthy for the range of chain parameters that were studied during the course of their simulations. A major focus of the work is comparison with Rouse model predictions for chain dynamics, but a considerable number of other parameters were also studied. These workers report an extended series of molecular dynamics simulations of 5, 10, and 20 kDa poly(ethylene oxide) ring polymers in the melt, corresponding to polymers having 120, 227, or 455 monomers. Simulations were made with a united atom force field[38,39] under isothermal/isobaric conditions, with $T=413$ K and $P=1$ atm. The force field parameters were expected to be sufficiently realistic that quantitative comparisons with experiments were expected to be possible, as confirmed in the paper. For the largest polymer, the simulation cell contained more than 50,000 atoms, the simulation being extended to an equivalent of 2.2 $\mu$S.

In considering Tsalikis, et al.’s findings on the applicability of the Rouse model to ring melts, one might say that the cup is half full or half empty. Tsalikis, et al., chose to emphasize points where their simulations clearly match Rouse model predictions. Here we emphasize the differences, points where the simulations do not match the Rouse model as applied to a ring polymer.

Tsalikis, et al., use their simulation data to compute for their rings the mean-square Rouse amplitudes. Their interpretation of a Rouse model for rings predicts that the normalized amplitude $\langle {\left(X_{p\alpha }\left(0\right)\right)}^2{p}^2/N$ should be independent of mode number p and polymer bead count N. This prediction is rejected by Tsalikis, et al.’s, simulations: The normalized amplitudes depend on p, and at small p are smaller than predicted by the Rouse model. For the $N=455$ polymer, the normalized amplitude for $p=2$ is modestly more than half its value for the same polymer at large p. The range of smaller p-values for which the normalized amplitudes are below their large-p limit increases with increasing N, the increase in the range being approximately linear in N. However, for all N studied, the normalized amplitudes do appear to go to the same large-p limit, so the model is arguable valid for large p.

For each of their chain lengths and $p=2,4,6,8,10,$ and 12, Tsalikis, et al., also report the time dependence of the time correlation functions $\langle X_{pi}\left(t\right)X_{pi}\left(0\right)\rangle$ . Figure 2 shows a sampling of their measurements (dots). The figure also shows our fits to stretched exponentials (solid lines) and to pure exponentials (dashed lines, obtained from fits to the initial slope). Here $\alpha$ and $\beta$ are fitting parameters. The correlation functions were normalized to unity at $t=0$. If Figure 2 is examined, it is apparent that the relaxation of $\langle X_{pi}\left(t\right)X_{pi}\left(0\right)\rangle$ is described well by a stretched exponential in time, except for a few of the largest-t points, contrary to the Rouse model prediction that the relaxations should be simple exponentials.

The stretched exponential is characterized by $\alpha$ and $\beta$. Numerical values for the fitting parameters and the average decay rate $\overline{\gamma }$ are seen in Table 2. For each molecular weight, $\alpha$ increases nearly 40-fold between $p=2$ and $p=12$. $\beta$ is close to unity for the $N=120$ polymer, but about 0.8 for the two larger rings.

Tsalikis, et al., calculated the normalized cross-correlations $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$, eqn. 10, between the Rouse amplitudes. They observe that the cross-correlations are not large; $|\langle X_{pi}\left(t\right)X_{qi}\left(0\right)\rangle |$ is almost always less than 0.1. Before considering this result, we ask how large $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$ is plausibly likely to be. If $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle =1$, modes p and q are perfectly cross-correlated; the value of one determines the value of the other. If one mode is cross correlated with several others, $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$ for any pair of modes must be considerably less than unity. For example, if a given mode is equally cross-correlated with n other independent modes, then at most the n modes completely determine the value of the given mode, in which case the cross-correlations would be of typical size $1/n$. One might also ask how accurately $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$ can be determined. If $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$ lies within simulational error of zero, non-zero values for $\langle X_{pi}\left(t\right)X_{qj}\left(0\right)\rangle$ are uninteresting.

Tsalikis, et al.,[13] evaluated the relaxation of the correlation function For a linear chain, $\mathbf{u}$ is the end-to-end vector. Ring polymers have no ends, so $\mathbf{u}$ is usefully defined to be a vector from a bead to another bead half-way around the ring. $\mathbf{u}$ has two paths to relaxation. First, its magnitude $\left|\mathbf{u}\right|$ fluctuates around its average value, contributing a relaxation; however, this process cannot relax the correlation function to zero. Second, as the dominant process, $C_{uu}\left(t\right)=\langle \mathbf{u}\left(t\right)·\mathbf{u}\left(0\right)\rangle$ relaxes by chain reorientation. At long times, $\mathbf{u}\left(t\right)$ and $\mathbf{u}\left(0\right)$ cease to be correlated, so their cross-correlation function decays to zero. As shown by the original authors, $C_{uu}\left(t\right)$ follows a stretched exponential in time, with an average relaxation time that increases as $N^{1.9}$, based on the three molecular weights studied. Tsalikis, et al., compare the functional form of $C_{uu}\left(t\right)$ that they obtained from their simulations with predictions from the Rouse model. The Rouse predictions for the time dependences, other than going to zero at long time, do not resemble the simulation determinations of the time dependence of $C_{uu}\left(t\right)$.

Other quantities studied by Tsalikis, et al.,[13] include the intermolecular and intramolecular atom-atom radial distribution functions, which had the expected forms. Static structure factors were calculated and found to be in good agreement with experiment. The distributions of end-to-end distances of chain segments of different lengths were calculated as functions of the length of the segments. The distributions were in general not described by Gaussians, especially for the larger rings. In contrast, an initial assumption of the Rouse and Kirkwood-Riseman models is that the end-to-end distances have Gaussian distributions, an assumption leading to the potential of average force between pairs of linked beads. If the bead-to-bead distances do not have Gaussian distributions, the potential of average force between them is not a Gaussian and does not correspond to a simple spring. Local dynamics were studied using the temporal autocorrelation functions of the torsion angles; the functions were described well with stretched exponentials in time. Finally, these authors asked how many other polymer chains a given chain typically interacts with. As a sensible approximation to this number, they calculated $K_1\left(r\right)$, the average number of other chains that had their center of mass within the radius of gyration of the chain of interest. For their ring polymers $K_1\left(r\right)$ was in the range 1.75-2.75. For linear chains having the same three molecular weights, $K_1\left(r\right)$ was in the range 8.5-9.5, with $K_1\left(r\right)$ increasing as the chain molecular weight was increased from 5 kDa to 20 kDa.

From the above, Tsalikis, et al., concluded that the Rouse model is valid for their systems.

Papadopoulos, et al.,[40] report united atom simulations of polyethylene oxide rings in the melt and in dilute solution in melts of three different linear polyethylene chains. Simulations used the modified TrAPPE force field[38,39] executed with GROMACS[41] held at T = 413 K and 1 atmosphere. Comparison was made with simulated melts of the three linear chains and with experimental studies by Goosen, et al.,[42] using nuclear spin echo spectroscopy. Goosen, et al., concluded that the segmental dynamics of dilute rings in a melt of linear chains were primarily determined by the dynamics of the host polymers. The ring polymers contained 456 monomers, for a molecular weight of 20 kDa, while the linear polymers had molecular weights of 1.8, 10, and 20 kDa, corresponding to chain N of 41, 228, and 456. Simulations included 8 rings and 72-720 linear chains with $>{10}^5$ atoms in a simulation cell.

In addition to a series of other static and dynamics properties, Rouse amplitudes $X_p\left(t\right)$ were used to compute $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ and $\langle X_p\left(0\right)X_p\left(t\right)\rangle$, the former for $N/p^2$ from 100 down to $<0.02$ and the latter for $p=2,4,\dots 12$. For small p, the Rouse model predicts $\langle {\left(X_p\left(0\right)\right)}^2\rangle \sim N/p^2$. For the ring melt and the blends, this result was confirmed for $N/p^2>1$. For larger p, i.e., $N/p^2<1$, $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ deviates downward from the small-p predicted value, attaining at the largest p examined perhaps half the value expected from an extrapolation of small-p behavior.

Papadopoulos, et al.’s determinations of the time correlation functions $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ appear in Figure 3. They report their determinations as smooth curves, appearing in the figure here as dotted lines. We fit to stretched exponentials (solid lines) and show simple exponentials (dashed lines) where appropriate. There is one behavior for the ring melt and for dilute rings in the 1.8 kDa chain melt (Figs. Figure 3a and b), and a somewhat different behavior for dilute rings in melts of the 10 and 20 kDa chains (Figs. Figure 3c and d).

Our description of the relaxation functions is not entirely the same as that of Papadopolous, et al. Numerical fits clarify issues visible in the figures. In the ring melt, and in dilute solution in the 1.8 kDa linear chains, $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ shows a stretched-exponential relaxation at earlier times, followed by a sharp transition to a simple-exponential relaxation at later times. The transition, which is especially prominent for $p=2$ and $p=4$, occurs at earlier times and smaller values of $\langle X_p\left(0\right)X_p\left(t\right)\rangle /\langle {\left(X_p\left(0\right)\right)}^2\rangle$ as p is increased. For larger p, the transition is more readily apparent in the solution of rings in the 1.8 kDa linear chain melt than in the ring melt. In contrast, for rings in dilute solution in the 10 kDa and 20 kDa melts, $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ for $p=2$ and for $p=4$ relaxes as a single stretched exponential out to the longest times observed. At larger p, $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ fluctuates around the fitted stretched exponential. The transition between short and long time behavior is not seen in many other studies, so it appears to be worthy of further investigation,

Papadopolous, et al., report integrated times ${\tau }_p$ for their four systems and the six smallest values of p. They report that ${\tau }_p$ scales approximately as ${(N/p)}^2$, ${\tau }_p$ being several-fold larger for rings in melts of the larger-N linear polymers than for rings in their own melts. Note, however, that Papadopolous, et al., found Gaussian distributions of distances between remote parts of the rings.

Kopf, et al.,[14] demonstrate a novel simulational test of the Rouse model. They consider systems containing 16 to 25 chains of 10 to 150 beads, in which the forces between all pairs of chains are exactly identical, but in which the beads on some or all of the polymer chains are made four or 100 times as massive as the original ’light’ beads. The forces between the beads were the FENE potential and a truncated, purely repulsive bead-bead Lennard-Jones potential. From basic statistical mechanics, a change in bead mass should have no effect on the static properties of the chains in a melt, an outcome that was confirmed simulationally. In mixtures, increasing the mass of the heavier beads slows down the motions of the light-bead chains. Kopf, et al., took advantage of the fact that they were doing molecular dynamics to calculate the velocity autocorrelation function through multiple oscillations out to long times. The frequency of the oscillations is relatively independent of the fraction of light or heavy polymers in the system, suggesting that the oscillations in the velocity autocorrelation function arise primarily from intramolecular interactions. The Rouse model remained accurate in light-heavy polymer mixtures. Rouse modes were found not to be cross-correlated. Rouse time correlation functions decayed approximately exponentially in time. Contrary to the Rouse model, these simulations observed subdiffusion (mean-square center-of-mass displacement proportional to $t^{0.8}$) on shorter time scales. A nominal entanglement time was used to estimate a nominal tube diameter, which the authors also described as a characteristic length for slowing down of monomer motion. Their results indicated that the nominal tube diameter is independent of the monomer mass, implying that the tube diameter is a static rather than a dynamic quantity, consistent with topological pictures for entanglements.

We turn to Paul, et al.,[43] who studied 40 to 120 chains of a C${}_{100}$ polyethylene using atomistic molecular dynamics. Their simulations included both an explicit-atom model and also a unified atom model in which each CH${}_2$ group was treated as a single atom. The polymer was chosen to be long enough that it could reasonably be expected to show Gaussian behavior for its static chain statistics, yet short enough that its dynamics would be expected to have Rouse-like and not reptational behavior. The authors recognized that the assumption of Rouse-like behavior in unentangled melts required examination. A stochastic dynamics simulation was used to equilibrate the samples, while data was obtained using molecular dynamics. Static behavior was tested by calculating the static structure factor; good agreement between simulation and experiment was found. In addition to other dynamic studies, large-scale dynamic behavior was compared with expectations from the Rouse model.

The end-to-end vector reorientation time and the long-time self diffusion coefficient are consistent with the same value for the segmental friction coefficient, these results being applicable ’on time scales larger than the Rouse time.’ However, contrary to the Rouse model, at times shorter than the Rouse time the center-of-mass diffusion is subdiffusive, being proportional to $t^{0.83}$ or so. Static mean-square amplitudes $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ of Rouse modes were calculated. For $p\le 3$, the small-p Rouse model expectation $\langle {\left(X_p\left(0\right)\right)}^2\rangle \sim p^{-2}$ was observed. For $p>3$, the mean-square mode amplitudes decrease approximately as $p^{-3}$, not as $p^{-2}$. The equal-time cross-correlation functions $\langle X_p\left(0\right)X_q\left(0\right)\rangle$ ($p\ne q$) were found to vanish to “...within the error bars in the simulation”; they interpret this finding to imply that ``...the Rouse modes are still good eigenmodes for a dynamic analysis...”, a claim that would also require that $\langle X_p\left(0\right)X_q\left(t\right)\rangle \ne 0$ for $p\ne q$ and $t>0$.

Paul, et al., also calculated the temporal autocorrelation functions $\langle X_p\left(t\right)X_p\left(0\right)\rangle$. A plot of correlation functions with $p=1,2,$ and 3 finds that the three correlation functions decay nearly exponentially as $exp(-\mathsf{\Gamma }{p}^{2}t)$, a single value of $\mathsf{\Gamma }$ sufficing for all three values of p, with small deviations over the first quarter of the decay. For $p>3$, the $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ are markedly non-exponential. When plotted against $p^{2}t$, with increasing p the $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ decay more rapidly. The authors conclude that the Rouse model ’…is at most applicable to a few largest scale eigenmodes.’ They do, however, note that the self-diffusion coefficient and the rotational diffusion coefficient can be described self-consistently in terms of a single segmental friction factor.

These results were extended by Paul, et al.,[23] who considered the single-chain intermediate structure factor $g^{\left(1\right)}(q,t)$ for unentangled polyethylene molecules in a melt, comparing results from neutron spin echo spectroscopy with results from atomistic and from united atom molecular dynamics simulations. They continued to study C${}_{100}$ polyethylene, because the polymer is short enough not to be entangled and long enough to have Gaussian chain statistics. The corresponding Rouse model has two parameters, namely a bond strength revealed by the average segment length b, and a monomer drag coefficient revealed by the chain center-of-mass diffusion coefficient D, the latter determined both experimentally and from each of the two sets of simulations. The simulation values for $g^{\left(1\right)}(q,t)$ for the explicit-atom and unified-atom simulations were in agreement with the experiments over a factor of 6 in q and two orders of magnitude in the scaled time $Dt$.

Having validated the accuracy of the simulations, Paul, et al., then used the simulations to calculate the Rouse amplitudes, their time autocorrelation functions, and the $g^{\left(1\right)}(q,t)$ implied by the Rouse modes. The $g^{\left(1\right)}(q,t)$ predicted by the Rouse model only agrees with the simulations for a limited range of q ($\le 0.14{\AA }^{-1}$) and times $\le 4$nS. At larger q and separately at longer times, Rouse-model predictions of $g^{\left(1\right)}(q,t)$ are significantly smaller than $g^{\left(1\right)}(q,t)$ from experiment or as calculated directly in the simulations. The authors note three marked deviations between the simulational results and the Rouse model: First, for times less than a time $\le {\tau }_R$, diffusion is found by the simulations to be subdiffusive, with exponent 0.83, rather than diffusive; in contrast, Rouse-model chains exhibit diffusive center-of-mass motion at all times. Second, simulations find that only the lowest Rouse modes, $p\le 3$, have relaxations that scale as $p^{2}t$; for larger p, Paul, et al., attribute deviations from $p^{2}t$ behavior to non-Gaussian chain statistics at short distances. Third, in the simulations each $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ decays as a stretched exponential in time; the $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ of the Rouse model all decay as pure exponentials.

Finally, $g^{\left(1\right)}(q,t)$ from the simulations, together with a Gaussian approximation was used to calculate a mean-square center-of-mass displacement $\langle {(\Delta R_{\mathrm{cm}}\left(t\right))}^2\rangle$. It should be emphasized that Doob’s theorem guarantees as a mathematical certainty that if the physical requirements leading to the Gaussian approximation are valid, then as a mathematical certainty $\langle {(\Delta R_{\mathrm{cm}}\left(t\right))}^2\rangle$ increases linearly in time. However, as found by Paul[43] at short times the calculated center-of-mass motion is subdiffusive, i.e., $\langle {(\Delta R_{\mathrm{cm}}\left(t\right))}^2\rangle$ grows as $t^{0.83}$ not as $t^1$. The Gaussian-approximation estimate of the mean-square displacement does agree with the simulations at long times $t\ge {\tau }_R$, at which the center-of-mass motion is diffusive. At times shorter than the Rouse time, $\langle {(\Delta R_{cm}\left(t\right))}^2\rangle$ as determined by the simulation is considerably larger than $\langle {(\Delta R_{cm}\left(t\right))}^2\rangle$ inferred from $g^{\left(1\right)}(q,t)$ and equation 23, showing that the Gaussian approximation is not valid in these systems at shorter times.

Several theoretical advances followed this work. Smith and Paul[44] used quantum chemistry calculations to generate an improved set of force parameters for simulations of 1,4-polybutadiene. Harnau, et al.[45,46] proposed that the experimental results of Paul, et al.[23] could be understood by replacing the Rouse model with a semiflexible chain model that takes into account chain stiffness. The semiflexible chain model with reasonable parameters agrees well with Paul, et al.’s experimental and simulational determinations of $g^{\left(1\right)}(q,t)$, of the mean-square displacement of the central monomer as a function of time, and of the Rouse-mode relaxation times.

Smith, et al.,[47] present simulations of an unentangled C${}_{100}$H${}_{202}$ 1,4-polybutadiene melt using a united atom, quantum chemistry-based potential, the focus of the work being to examine the presence of non-Gaussian displacement distributions of polymer beads in a melt. The single-chain intermediate structure factor $g^{\left(1\right)}(q,t)$ was determined from neutron spin-echo measurements and separately from molecular dynamics simulations using Smith and Paul’s[44]united atom potential. For $0.05\le q\le 0.30$ Å${}^{-1}$ and times out to 17 nS, experimental and simulated values of $g^{\left(1\right)}(q,t)$ were in good agreement. The center-of-mass motion was diffusive at long times but subdiffusive ($\langle {\left(\delta R\left(t\right)\right)}^2\rangle \sim t^{0.8}$) at times shorter than ${\tau }_R\approx 15$ nS. The simulated $g^{\left(1\right)}(q,t)$ was compared with predictions of the Rouse model and several of this model’s proposed modifications, finding that none of the models reproduced the simulations. Simulations were also use to calculate $\langle X_p\left(t\right)X_q\left(0\right)\rangle$, the correlation function vanishing for $p\ne q$, at least for $p,q\le 4$. Use of the simulated $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ in the Rouse form for $g^{\left(1\right)}(q,t)$ also did not lead to agreement of this modified Rouse model with experiment.

The authors observe that the Gaussian approximation for $g^{\left(1\right)}(q,t)$ is only appropriate if the distribution of relative bead displacements ${\mathbf{R}}_m\left(t\right)-{\mathbf{R}}_n\left(0\right)$, ${\mathbf{R}}_m\left(t\right)$ being the position of bead m at time t, is Gaussian at all times. To examine the consequences of this observation, they calculated $g^{\left(1\right)}(q,t)$ using the Gaussian approximation and values of mean-square displacements $\langle {({\mathbf{R}}_m\left(t\right)-{\mathbf{R}}_n\left(0\right))}^2\rangle$ obtained from their simulations, finding that this calculated $g^{\left(1\right)}(q,t)$ was in good agreement with $g^{\left(1\right)}(q,t)$ as predicted by the Rouse model, but did not agree with $g^{\left(1\right)}(q,t)$ as calculated directly from the simulation, thus showing the importance of non-Gaussian particle displacements. Smith, et al., conclude that the non-Gaussian distribution of bead displacements ${\mathbf{R}}_m\left(t\right)-{\mathbf{R}}_n\left(0\right)$ is responsible for the observed failure of the Rouse model in polymer melts.

Two sorts of non-Gaussian behavior possible here. The first is that the distribution of displacements $P(\Delta {\mathbf{R}}_m(\Delta t))$ for each bead separately could be non-Gaussian. The second is that the distributions of displacements of pairs of beads could be cross-correlated. Thanks to the fluctuation-dissipation theorem, this latter possibility is equivalent to the statement that there are significant hydrodynamic interactions in polymer melts, a possibility that would only be surprising if the Rouse model were correct in polymer melts.

Harmandaris, et al.,[48], made atomistic simulations of 24-, 78-, and 156-atom (mean length) linear polyethylene melts, finding a diffusion coefficient D as well as the time autocorrelation functions of the polymer end-to-end vector and the Rouse mode amplitudes. Each autocorrelation function may be said to have a characteristic time $\tau$. The study was novel in that the authors deliberately simulated polydisperse melts having polydispersity index near 1.09. From these quantities, nominal monomer friction factors were extracted. Initial chain configurations were equilibrated using the end-bridging Monte Carlo scheme[49]. Molecular dynamics were executed using a sixth-order predictor-corrector model. The objectives of the study were to test the Rouse model, and to take advantage of the polydispersity to examine the dynamics of chains having different lengths, all in the same melt. Potential energies included a Lennard-Jones potential between non-bonded atoms, bond-bending and torsional potentials, and a Fixman potential[50] to keep bond lengths constant. The 24- and 78-atom carbon models were simulated in both the NVE and NVT ensembles; results agreed. The mean-square end-to-end distance $\langle {\left(\mathbf{r}\left(0\right)\right)}^2\rangle$, radius of gyration, and intermolecular bead-bead static correlation functions from the molecular dynamics simulation and the end-bridging Monte Carlo simulation were found to agree, confirming the validity of the two simulations. Local dynamics as estimated with the torsion angle temporal autocorrelation function showed that local dynamics become slightly slower as the chain length is increased.

Harmandaris, et al.,[48] calculated properties of the Rouse amplitudes $X_p\left(t\right)$. The mean-square static amplitudes $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ decrease with increasing p, for the first two modes with the $p^{-2}$ dependence predicted by the Rouse model, and at larger p much more rapidly, approximately as $p^{-3}$. For $p=10$, the discrepancy between the simulation and the Rouse model approaches an order of magnitude. The temporal correlation functions $\langle X_p\left(0\right)X_p\left(t\right)\rangle$, at least for the 83- and 117-carbon chain systems, also do not agree with the Rouse model, namely they are not simple exponentials, and their relaxation times do not scale with time as $p^{2}t$. On the other hand, for the end-to-end vector, the relaxation time of $\langle \mathbf{r}\left(0\right)·\mathbf{r}\left(t\right)\rangle$ calculated using the Rouse model and the relaxation time inferred from simulations of $\langle X_1\left(0\right)X_1\left(t\right)\rangle$ agree well. From observations of the chain center-of-mass motion over long times, a chain diffusion coefficient D and therefore a monomer friction factor $\zeta$ can be inferred. Contrary to the Rouse model, $\zeta$ depends on chain length, increasing threefold from the shortest to the longest chains studied. However, $\zeta$ appears to reach an asymptotic value for the longer chains, those with $N>60$ or so. Harmandaris, et al., propose that $N=60$ is therefore a lower limit below which the Rouse model is inapplicable. These workers also calculated, from the diffusion coefficient, a zero-shear viscosity. The calculation was based on Rouse’s theory, which in some other respects does not describe the dynamics of these systems, but the calculated viscosity determined by ${\tau }_1$ and the $p=1$ mode does approximately follow Rouse model predictions.

Krushev, et al.,[51] simulated melts of 1,4-polybutadiene. Their interest was to determine the effects of torsion barriers on molecular motions. To do this, they examined the static structure factor, Rouse mode amplitudes, Rouse-Rouse temporal autocorrelation functions, and the intermediate scattering function $g^{\left(1\right)}(q,t)$. Their polymer melts incorporated 40 polymer chains, each with 29 or 30 subunits, all with united atom potentials, including a model with chains incorporating vinyl groups, a model with chains not incorporating vinyl groups, and a model with no vinyl groups and all rotational potentials set to zero. The three models have the same distribution for their radii of gyration.

Rouse mode amplitudes had at most weak cross-correlations, $\langle X_p\left(0\right)X_q\left(0\right)\rangle$ for $p\ne q$ being less than 2% of $\langle {\left(X_p\left(0\right)\right)}^2\rangle$. The mode amplitudes $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ only followed the Rouse prediction $\langle {\left(X_p\left(0\right)\right)}^2\rangle \sim p^{-2}$ for $p<5$. At larger p, the mean-square amplitude depended to good accuracy as $p^{-3}$, which is the prediction for a freely rotating polymer.[52] The calculated static structure factor was not affected by adding or deleting the torsion potential. The time correlation functions $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ showed stretched-exponential, not single-exponential, behavior for all p that were studied over an adequate time range. Time correlation functions with and without torsion potentials were very nearly the same when plotted in reduced time units in which the $1/e$ time was identified as unit time, showing that chain stiffness arising from torsion potentials did not have a significant effect on the form of the relaxation of $X_p\left(t\right)$.

The intermediate scattering function $g^{\left(1\right)}(q,t)$ also did not follow the Rouse model predictions, namely the Rouse model predicts an over-rapid decay of $g^{\left(1\right)}(q,t)$ at larger q, and underpredicts the degree of stretching of $g^{\left(1\right)}(q,t)$. $g^{\left(1\right)}(q,t)$ is not significantly changed when torsional potentials are added or removed from the potential energy; the authors infer that observed deviations from Rouse behavior are not due to internal rotation barriers. The Rouse model calculation of $g^{\left(1\right)}(q,t)$ agrees with the $g^{\left(1\right)}(q,t)$ calculated from the simulation on using the Gaussian approximation and the measured mean-square displacement. Neither calculation agrees with the actual $g^{\left(1\right)}(q,t)$. Furthermore, the center of mass mean-square displacements are subdiffusive at short times. The authors conclude that the Rouse model assumption – that atomic motions are described by a joint Gaussian random process – is thus shown to be incorrect.

Padding and Briels[53] report simulations of twelve chains of a C${}_{120}$H${}_{242}$ polyethylene melt. They simulated four different starting points for their melt, molecules having united atom potential energies. The potential energy used by these authors was not the simple-harmonic bond-length potential of the Rouse model. Bond lengths and angles had harmonic potential energies; a torsion potential energy was present; unbonded (separated by four or more atoms in a single molecule) pairs of united atoms have a Lennard-Jones potential. A weak coupling to a bath held the temperature fixed. Melt starting chain configurations were created by gradual compression of a dilute system in which only repulsive interatomic forces were present.

The time-dependent dynamic parameters that they obtained from their simulations include mean-square displacements, the end-to-end vector time autocorrelation function, the intermediate structure factor, and the stress tensor. A single set of numerical parameters for the number of segments N in a chain, for the diffusion coefficient D, and for the longest relaxation time ${\tau }_1$ described most of the dynamic quantities that they calculated, but only over distances longer than a limiting length scale,. This paper actually tested the relationships between N, D, ${\tau }_1$, and the calculated dynamic parameters, as predicted by the Rouse model, but did not test the Rouse model itself, Rouse’s description of the internal dynamics of a polymer chain.

Here the stress tensor was calculated as where ${\mathbf{v}}_i$ is the velocity of atom (or center of mass) i, ${\mathbf{r}}_i$ is the position of atom (or center of mass) i, and ${\mathbf{F}}_{ij}$ is the force exerted on atom (molecular center of mass) i by atom (molecular center of mass) j. The forces between two molecules can create a torque, an antisymmetric part of the stress tensor, on each molecule. The stress tensor was identified as leading to the zero-shear relaxation modulus via where $\mathbf{P}$ is the symmetrized traceless part of $\sigma$.

Padding and Briels concluded that there is a shortest length scale $b\approx 1.2$ nm over which the Rouse model is valid in their simulations. The length scale manifests itself as the shortest Rouse-mode wavelength for which the model works, the shortest distance over which a mean-square displacement must occur before Rouse behavior is seen, and the shortest wavelength for which the simulated $g^{\left(1\right)}(q,t)$ agrees with the model. Over shorter distances and times, matters became more complicated. Padding and Briels calculate $g^{\left(1\right)}(q,t)$ for a series of wave vector magnitudes. At small q, only center-of mass diffusion is seen. At larger q, $g^{\left(1\right)}(q,t)$ has contributions from polymer internal modes. At larger q, $g^{\left(1\right)}(q,t)$ from the simulation and $g^{\left(1\right)}(q,t)$ calculated from the Rouse modes and the diffusion coefficient found at small q do not agree, $g^{\left(1\right)}(q,t)$ from the Rouse formula decaying faster at long times than $g^{\left(1\right)}(q,t)$ from the simulations, the discrepancy becoming larger at larger q. The short-distance internal chain modes are thus not the same as the internal modes predicted from the Rouse model. Similarly, measurements of mean-square displacements from simulations only agree with the Rouse model when mean-square displacements are greater than (1.1 nm)${}^2$. Finally, at short times the simulated shear relaxation modulus $G\left(t\right)$ “...does not behave Rouse-like at all...”, but corrections due to this issue at long times were limited in size.

In a further paper, Padding and Briels[54] report simulations of a heavily coarse-grained (one bead = 20 monomers, 120 chains in a simulation box) C${}_{120}$ linear polyethylene that incorporated a complicated switchable scheme for enforcing chain uncrossability. The scheme could be turned off, leading to simulations of a melt in which polymer chains could pass through each other. Interactions included non-bonded, bonded, and bending-angle contributions to the bead-bead potential of average force. Beads incorporated as many monomers as feasible without making the beads larger than a nominal tube radius. Padding and Briels calculated the mean-square displacements, both of single blobs and of the chain centers of mass. For the unentangled chains, mean-square displacements increased linearly with time. Adding the uncrossability constraint reduced the mean-square displacements and gave them a sublinear time dependence over a considerable time regime.

Padding and Briels[54] calculated the time-dependent Rouse amplitudes and evaluated their time correlation functions. Fits were then made to stretched exponentials in time. Chains had six beads, so they only had five internal Rouse modes. When chain crossing was permitted, the time correlation functions were very nearly single exponentials. Adding chain crossing constraints and a bond bending potential led to appreciably non-exponential relaxations, the stretching parameter $\beta$ falling from close to unity in the absence of chain crossing constraints or a bond bending potential to 0.77 for the three highest modes when the constraint and potential were added. The chain crossing constraint considerably increased the relaxation times of the $p=1$ and $p=2$ modes but did not increase substantially the relaxation times of the three higher-p modes. Padding and Briels determined effective relaxation times ${\tau }_p^{\mathrm{eff}}$ for their five modes. However, instead of calculating ${\tau }_p$ from the stretched-exponential fitting parameters, the authors did numerical integrals of the measured $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ curves. The Rouse relaxation rates, equation 16, were evaluated for each p. In the Rouse model, $W_p^{\mathrm{eff}}$ is independent of p. In the presence of chain stiffness, and more dramatically in the presence of uncrossability, $W_p^{\mathrm{eff}}$ was found to increase several-fold as p was increased, the major change from the Rouse model being that in the presence of chain stiffness and uncrossability $W_p^{\mathrm{eff}}$ is reduced for small p. Finally, Padding and Briels[54] calculated the system’s intermediate structure factor $g^{\left(1\right)}(q,t)$ for a series of values of q. At small q, $g^{\left(1\right)}(q,t)$ is relaxed by whole-body translational diffusion, a fit giving the polymer’s diffusion coefficient. In the presence of chain uncrossability and larger q, $g^{\left(1\right)}(q,t)$ did not agree with the Rouse model predictions.

Padding and Briels[20] further extended their work on polyethylene by making extended united atom simulations of melts of seven different polyethylenes, using 80 to 180 chains with 80 to 1000 carbon atoms coarse-grained into 4 to 50 blobs at 450 K and a density 0.761 g/cm${}^3$. They stress that the eliminated internal coordinates become thermal bath variables, and contribute to the motion of the blobs as unseen random thermal forces and a friction factor, which they treated as a scalar with no associated memory function. The paper considers a considerable list of different dynamic parameters. This chapter is only concerned with the behavior of the Rouse amplitudes. They calculated the Rouse amplitudes and their time correlation functions. On plotting $8N{sin}^2(p\pi /2N)\langle {\left(X_p\left(0\right)\right)}^2\rangle$ against $N/p$, values of $8N{sin}^2(p\pi /2N)\langle X_p{\left(0\right)}^2\rangle$ for all chain lengths N superpose, but contrary to the Rouse model $8N{sin}^2(p\pi /2N)\langle X_p{\left(0\right)}^2\rangle$ is not independent of $N/p$; it instead falls off from slightly more than 3 to slightly more than 1 as $N/p$ is reduced, i.e., as p is increased. They interpret this non-Rouseian behavior as arising from their angular potential, which leads to some degree of chain stiffness.

Padding and Briels[20] also calculated the Rouse-Rouse time correlation functions. Rouse modes at time zero are not cross-correlated; they did not report what happens at later times. Rouse-mode temporal autocorrelation functions were found to decay as stretched exponentials in time. The stretching parameter $\beta$ was close to 0.7 near $N/p=1$. $\beta$ decreased to 0.55 or so near $N/p=2$, and then increased to near 0.7 at $N/p=6$. At larger $N/p$, $\beta$ was nearly constant. The dependence of $W_p^{\mathrm{eff}}$ on $N/p$ was examined. For $N/p\approx 1$, $W_p^{\mathrm{eff}}$ is independent of $N/p$. For $N/p$ modestly above 1 and out to 3 or so, $W_p^{\mathrm{eff}}\sim {(N/p)}^{-2}$. For $N/p>3$, $W_p^{\mathrm{eff}}\sim {(N/p)}^{-1}$. Padding and Briels note, for the second and third regimes, that these dependences are not in agreement with either the Rouse or the reptation model. Padding and Briels then propose that at large times $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ switches over from a stretched-exponential to a simple-exponential time dependence.

Abrams and Kremer[55] studied a bead and spring polymer melt, the interest being the effects of varying the equilibrium bond length ${\ell }_0$ relative to a nominal bead diameter $d_0$. In different simulations, the bead length was given 13 values in the range $0.73\le {\ell }_0/d_0\le 1.34$. The model contained 80 freely-jointed chains, each having 50 beads, at density 0.85 ${\sigma }^{-3}$ and nominal temperature $T=1$ in natural units, so that $d_0=\sigma$. Bonded beads were linked with a harmonic potential $\frac{k}{2}{(\ell -{\ell }_0)}^2$, ℓ being a bond length, with k and ${\ell }_0$ being simulational parameters. Non-bonded beads interacted with a truncated Lennard-Jones potential. The authors studied the time correlation functions of the Rouse amplitudes $X_p\left(t\right)$ and the mean-square displacements of the bead centers-of-mass.

Semi-log plots of the normalized $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ as functions of a normalized time $t{p}^2/N^2$ were presented for ${\ell }_0/d_0$ equaling 0.79, 0.97, and 1.24 and $p\le 5$. For ${\ell }_0/d_0=0.79$ , the plots were nearly linear. The curvature increased with increasing ${\ell }_0/d_0$. For the smallest ${\ell }_0/d_0$, plots of $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ for the different values of p nearly superpose. For larger ${\ell }_0/d_0$, the curves spread out modestly from each other, though the dependence on p is hard to discern. Abrams and Kremer extracted ${\tau }_p$ from fits of $exp(-t/{\tau }_p)$ to the early parts of the $\langle X_p\left(t\right)X_p\left(0\right)\rangle$ curves, and advanced from there to nominal friction factors ${\zeta }_p$. In doing this, they indicated that “…The Rouse model best describes the early phase of the decay of$\langle X_p\left(t\right)X_p\left(0\right)\rangle$ …", and describe behavior at later times as “…subtle non-Rouse-like behavior …" whose examination was deferred to a future study. The inferred ${\zeta }_p$ values were presented as averages over p. As a function of ${\ell }_0/d_0$, the averaged ${\zeta }_p$ increase rapidly at larger ${\ell }_0/d_0$. Abrams and Kremer also calculated the average number of other polymer beads within a distance $2^{1/6}\sigma$ of a bead of interest. That number increases, roughly from 0.4 to 1, over the observed range of ${\ell }_0/d_0$.

Doxastakis, et al.[56] report extensive atomistic and unified-atom simulations of melts of very short (40-115 atom) polyisoprenes, including 64 chains of C${}_{80}$ polyisoprene, and compare with measurements from ${}^{13}$C NMR, quasielastic neutron scattering, the torsional correlation function from the simulation, dielectric relaxation spectroscopy, and polymer self-diffusion. Because these authors did atomistic simulations, their simulations determined single-bond and few-atom motions that could be compared with ${}^{13}$C NMR and neutron scattering. Reasonable agreement between simulation and experimentally measured quantities, within the expected limits of accuracy of the simulations, was obtained. Dielectric relaxation measurements were interpreted in terms of a Kohlrausch-Williams-Watts function for a higher frequency peak and Rouse normal modes for a lower-frequency peak. The Rouse fit showed some deviation from experiment at higher frequencies. The mean-square amplitudes $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ of the Rouse modes only followed the theoretical $p^{-2}$ scaling for the first two or three modes; for larger p the measured amplitudes are smaller than that theoretical prediction. Plots of the simulated $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ against $p^{2}t$ should collapse onto a single line. If the $t=0$ amplitude is normalized out, the plots come respectably close to doing so. However, at short times $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ from the simulations fell well below a fit of the long-time $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ to a single exponential, especially for larger p. The simulated time autocorrelation function for the chain end-to-end vector is at early times also smaller than expected from the Rouse model. Finally, on uniting the various theoretical and fitted treatments of chain end-to-end relaxation, very good agreement is obtained between the theoretical form and the simulations. The authors conclude that the Rouse model is sustained by their simulations, for the quantities that they analyzed, a conclusion that neglects the issues they faithfully reported with the mode mean-square amplitudes.

Tsolou, et al.[11,57,58] report a series of molecular dynamics simulations of polybutadiene and polyethylene. Their first paper[57] simulated cis-1,4-polybutadiene based on a united atom description in which hydrogen atoms were merged with the carbon atom to which they were bonded. Bonds were represented as Hookian springs having a finite rest length; bend and torsion angles had associated potential energies. Non-bonded atoms interacted with a non-truncated Lennard-Jones potential. Melt simulations were done on monodisperse polymers having N of 32 to 400 carbon atoms in systems containing more than 10,000 united atoms for times out to 600 nS. End-bridging Monte Carlo methods were used to create rapid equilibration; simulations were based on multiple-time-step molecular dynamics. The system was thermostatted to constant temperature and pressure. A long series of static quantities were calculated, including the mean-square radius of gyration, mean-square end-to-end distance, characteristic ratio, specific volume as a function of chain length, density as a function of temperature, the intermolecular pair radial distribution function, and the static structure factor. For the last of these, the locations of the hydrogen atoms had to be backed out from the unified atom description.

Tsolou, et al.,[57] also calculated dynamic quantities, including the time autocorrelation functions for the torsion angles and the chain end-to-end vector ${\mathbf{R}}_E\left(t\right)$. Efforts to fit $\langle {\mathbf{R}}_E\left(t\right)·{\mathbf{R}}_E\left(0\right)\rangle$ as a sum of Rouse modes were unsatisfactory; on the other hand, $\langle {\mathbf{R}}_E\left(t\right)·{\mathbf{R}}_E\left(0\right)\rangle$ was fit accurately with a single stretched exponential in t. The nominal relaxation times from these fits increased with increasing polymer length N, namely $\tau \sim N^a$ with a increasing from 2.1 for the shortest polymers to 2.8 for the longest polymers. The change in a with increasing N was not obviously discontinuous.

Tsolou, et al., also determined the mean-square displacements of the chain centers-of-mass as functions of time, and inferred from the displacements the diffusion coefficient D. D depends on N approximately as a power law. Curiously, the slope of $log\left(D\right)$ against $log\left(N\right)$ is shallowest for intermediate values of N.

An algorithm was used to obtain a nominal primitive path for each polymer chain at a series of times. The primitive path from this algorithm is a smooth curve that follows the atomistic backbone. The diameter a of the corresponding tube is $64\text{Å}$, which is considerably larger than the experimental $38\text{Å}$ tube diameter reported[59] for the same system. a is much larger than the distance between neighboring polymer chains, showing that when a polymer chain attempts to move transversely to its primitive path, and encounters another polymer chain, in general it is able to continue to move in the same direction over considerable distances. The authors also computed the mean-square displacements $g\left(t\right)$ of the central beads of each chain. $g\left(t\right)$ appears to be a smooth curve that could be described as having sections that follow power-laws $t^{\alpha }$. However, $\alpha$ was never less than 0.4, so it never reached the $\alpha =0.25$ of the Rouse model. For $N=400$, the transition from an initial $\alpha =0.5$ down to $\alpha =0.4$ occurred at $\approx {10}^3$ pS.

Calculations of the single-chain intermediate structure factor $g^{\left(1\right)}(q,t)$ were made. The authors concluded that is no q for which $g^{\left(1\right)}(q,t)$ agrees with a fit to the Rouse model, including times shorter than a few nS. According to the authors, the tube-reptation model indicates that at times shorter than a few nS polymer chains are supposed to be performing Rouse-like motion, because they have not yet having encountered the walls of their tube. However, $g^{\left(1\right)}(q,t)$ obtained from these simulations decays more slowly than does $g^{\left(1\right)}(q,t)$ predicted by the Rouse model, using Rouse times calculated by fitting independently to the time correlation functions of the polymer end-to-end vector. The tube-reptation prediction does not match the simulations. Nonetheless, the authors were able to extract a nominal friction factor $\zeta$ from the Rouse form for the diffusion coefficient, even though the Rouse model does not appear to describe the dynamics.

Tsolou, et al.,[58] examined Rouse amplitudes and dynamic structure factors for simulated cis-1,4-polybutadiene melts. The simulations viewed 32 chains of a C${}_{128}$ polymer as functions of the system’s temperature and pressure, at temperatures from 165 to 413 K and pressures from one atmosphere to 3.5 kbar. Simulation methods duplicated those in earlier papers by Tsolou, et al.[60,61]. The authors first examined the time autocorrelation functions $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ of the Rouse mode amplitudes for p having various values in the range $(1,64)$. The $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ were all found to decay as stretched exponentials in time, not the exponentials predicted by the Rouse model, leading to a set of values for ${\tau }_p$ and ${\beta }_p$. The stretched exponentials were also characterized via their total correlation times

Tsolou, et al.[58] found that the $t_p$ depend on temperature via a modified Vogel-Fulcher-Tamman equation Here T is the absolute temperature, $t_{p0}$ and $D_p$ are fitting parameters, and $T_o$ is a characteristic temperature. Over the range of temperatures that were examined in the simulations, the temperature dependences of the $t_p$ do not depend markedly on p. Pressure dependences of the $t_p$ were obtained at temperatures 310 and 413 K. The $t_p$ increase exponentially with increasing pressure P. Viewed graphicly, the effect of P on $t_p$ does not depend a great deal on p. The authors define an activation volume for the $t_p$ via $\Delta V_p$ decreases roughly by two-fold between $p=1$ and $p=64$, and for smaller p is modestly smaller (by less than 10%) at the higher than at the lower temperature.

Tsolou, et al.[58] also calculated the single-chain intermediate structure factor[57,62] in which n and m label two of the N segments of a single chain, q being the magnitude of the scattering vector and $R_{nm}\left(t\right)$ being the distance between segments n and m at two times separated by t. $g^{\left(1\right)}(q,t)$ was found to be described by a stretched exponential in t. The stretching exponent $\beta$ was reported to change with pressure and to decrease with increasing q. The total correlation times for $g^{\left(1\right)}(q,t)$ were found to decrease with increasing q and to increase exponentially with increasing pressure. The activation volumes $\Delta V\left(T\right)$, as calculated from the pressure dependences of these total correlation times, decrease with increasing q.

Tsolou, et al.[58] calculated the single-chain incoherent scattering factor $g^{\left(1s\right)}(q,t)$, which differs from $g^{\left(1\right)}(q,t)$ in that in equation 29 the restriction $n=m$ is forced. $g^{\left(1s\right)}(q,t)$ decays as a stretched exponential in time. $\beta$ was found to increase from $\approx 0.5$ to $\approx 0.9$ as q was increased over the observed range. The total correlation time $t_c$ decreased strongly with increasing q. Lines to guide the eye, drawn as $\tau \sim q^{-2/\beta }$ for smaller q and $\tau \sim q^{-2}$ at larger q, are harmonious with the q dependences of $t_c$ at multiple temperatures and a full range of pressures. The peculiar $-2/\beta$ exponent is an artifact of the stretched-exponential form $exp(-{(t/\tau )}^{\beta })$ used to parameterize $g^{\left(1s\right)}(q,t)$. If the parameterization had instead been the rational alternative then for smaller q the result would have been $\mathsf{\Gamma }\sim q^2$, while for larger q the form $\mathsf{\Gamma }\sim q^{2\beta }$ would have appeared as an approximant.

The Rouse model predicts that the relaxation time of the $p^{\mathrm{th}}$ mode should depend on p as $p^{-2}$. As the model also predicts that modes relax as simple exponentials in t, not the stretched exponentials actually found, there is no theoretical basis for identifying the total correlation time with the Rouse time. Indeed, $p^2{t}_p$ is not independent of p; it instead decreases by about 30% as p is increased from 1 to 20. Tsolou, et al.,[58] used this observation to estimate a longest relaxation time and hence a model-dependent zero-shear viscosity for the system. The inferred viscosity follows a Vogel-Fulcher-Tamman form.

Tsolou, et al.,[11] simulated melts of ring polyethylenes containing 24-400 carbon atoms per ring at nominal temperature and pressure of 450 K and 1 atmosphere; simulation boxes contained 3000-20,000 united atoms. Simulations were made with a united atom model treating methylene units as single atoms, a harmonic bond-stretching potential, a harmonic-in-bond-angle bending potential, a bond-torsional potential, and a 12-6 Lennard-Jones potential for atoms separated by more than three bonds, using the r-RESPA algorithm[63] for molecular dynamics. As two of a large number of properties (most not considered in this review), they calculated $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ and $\langle X_p\left(0\right)X_p\left(t\right)\rangle$. For $N/p^2\ge 2$, the mean-square amplitude $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ increased linearly in $N/p^2$. At smaller $N/p^2$, the increase in $\langle {\left(X_p\left(0\right)\right)}^2\rangle$ with increasing $N/p^2$ was much more rapid than linear in $N/p^2$, contrary to any prediction of linear behavior at all $N/p^2$. The time correlation functions $\langle X_p\left(0\right)X_p\left(t\right)\rangle$ were found to depend on t as stretched exponentials in time, not the simple exponentials of the Rouse model. Comparison was made between a Rouse prediction $\tau \sim N^2/p^2$ for small p, and the ${\tau }_p$ calculated from the time integral of $\langle X_p\left(0\right)X_p\left(t\right)\rangle$. The prediction was sustained for $N/p\ge 30$. At smaller $N/p$, ${\tau }_p$ decreases two-fold as $N/p$ is reduced from 30 toward 1. Rouse model predictions for the time correlation function and its zero-time value are therefore confirmed only for larger values of $N/p$.

Bulacu and van der Giessen[64] simulated the effect of bending and torsional potential energies on a polymer melt. Their systems included up to 1000 chains with $5\le N\le 250$. Their polymer was a bead-spring model, with a 6-12 Lennard-Jones potential truncated at its minimum, bonds between beads represented with a FENE potential, a cosine harmonic bending potential with ${\theta }_0=109.5^o$, and a coupled bending-torsion potential The torsion potential refers four beads in a row along a polymer, the two internal angles ${\theta }_{i-1}$ and ${\theta }_i$ formed by the two overlapping sets of three beads in a row, and the dihedral angle $\varphi$. The $a_n$ were determined by quantum calculations for n-butane as $(a_0,a_1,a_2,a_3)=(3,-5.9,2.06,10.95)$. Simulations used the velocity-Verlet algorithm; temperature was held steady with a heat bath’s random force and friction factor. $k_{\theta }$ and $k_{\varphi }$ are stiffness coefficients.