Non-Oscillatory Central Schemes for a Traffic Flow Model with Arrhenius Look-Ahead Dynamics

Non-Oscillatory Central Schemes for a Traffic Flow Model with Arrhenius Look-Ahead Dynamics Alexander Kurganov and Anthony Polizzi Abstract We develop non-oscillatory central schemes for a traffic flow model with Arrhenius lookahead dynamics. This model, proposed in [A. Sopasakis and M.A. Katsoulakis, SIAM J. Appl. Math., 66 (2006), pp. 92 944], takes into account interactions of every vehicle with other vehicles ahead ( look-ahead rule) and can be written as a one-dimensional scalar conservation law with a global flux. The proposed schemes are extensions of the first-order staggered Lax-Friedrichs scheme and the second-order Nessyahu-Tadmor scheme, which belong to a class of Godunov-type projection-evolution methods. In this framework, a solution, computed at a certain time, is first approximated by a piecewise polynomial function, which is then evolved to the next time level according to the integral form of the conservation law. Most Godunov-type schemes are based on upwinding, which requires solving (generalized) Riemann problems. However, no (approximate) Riemann problem solver is available for conservation laws with global fluxes. Therefore, central schemes, which are Riemann-problem-solver-free, are especially attractive for the studied traffic flow model. Our numerical experiments demonstrate high resolution, stability, and robustness of the proposed method, which is used to numerically investigate both dispersive and smoothing effects of the global flux. Introduction The main goal of this paper is to develop a reliable and robust numerical method for a new deterministic traffic flow model recently proposed in [25]. Traffic flow has been modeled by PDEs since the pioneering works [4] and [2]. The main idea in these models is to view cars as moving particles and to extend fluid dynamics approaches to traffic flows. We refer the reader to [6, 7, 8, 23, 26] for examples of gas-kinetic-type (mesoscopic) models and to [6, 8, 27] for some hydrodynamics-like (macroscopic) ones. Traffic flow has been also extensively modeled by stochastic differential equations (see [24, 25] and the references therein), optimal velocity models (see, e.g., [, 5, 20]), and cellular automaton models (see, e.g., [2, 6, 7]). The simplest deterministic continuum traffic flow model is ([4, 2]): u t + f(u) x = 0, f(u) = uv (u), V (u) = V m ( u), (.) Mathematics Department, Tulane University, New Orleans, LA 708, USA; kurganov@math.tulane.edu Tulane University, New Orleans, LA 708, USA; apolizzi@tulane.edu

2 A. Kurganov and A. Polizzi where u(x, t) is a car density (measured in cars per car length so that 0 u ), V m is the maximum possible speed, and V (u) is a dimensionless velocity function that is assumed to depend only on the density. In [25], this model has been extended to a more realistic one by taking into account interactions with other vehicles ahead ( look-ahead rule). This is done by treating vehicles movement as an asymmetric exclusion process (ASEP) on a periodic lattice representing a one-lane highway partitioned into N cells, L = {, 2,..., N}. Vehicles are represented as a spin configuration σ, which is defined for each x L as {, if a vehicle occupies cell x, σ(x) = 0, if the cell is empty, and offers the interpretation of the state of the system. Letting σ = {σ(x) : x L} gives a state variable within space {0, } L. This approach models random dynamics in traffic, as the spin configuration is treated as a stochastic process. The effect cars have on each-other is represented in the interaction potential of vehicles in the ASEP model, in which the Arrhenius look-ahead dynamics is incorporated. This leads to a stochastic model presented and studied in [25]. It is also shown in [25] that under certain mild assumptions, the stochastic process describing traffic movement can be treated as deterministic. To this end, the fluctuations in {σ(x), x L} are approximated by the law of large numbers, and the resulting PDE is the following scalar conservation law with a global flux (see [25] for details): u t + F (u) x = 0, F (u) = V m u( u) exp( J u), (.2) where the contribution of short range interactions to the flux is modeled by the convolution Here, and therefore, J w(x) := J(r) := γ ϕ ( r γ J(r) = x J(y x)w(y) dy. (.3) ) {, 0 < r <,, ϕ(r) := 0, otherwise, { /γ, 0 < r < γ, 0, otherwise, where γ is a positive constant proportional to the look-ahead distance. Introducing the antiderivative, U := x u(ξ, t) dξ, equation (.2) (.3) can be rewritten as: ( ) U(x + γ) U(x) u t + F (u, U) x = 0, F (u, U) = V m u( u) exp. (.4) γ We are interested in developing an accurate and reliable numerical method for equation (.4). Since this model is an extension of the simplest model (.), one may want to apply methods designed for scalar hyperbolic conservation laws to the global flux equation (.4). The most popular methods are finite-volume methods (see, e.g., [4, 9, 2]) and, in particular, Godunovtype schemes. These schemes form a class of projection-evolution methods, in which a computed

Central Scheme for a Traffic Flow Model 3 solution is first interpolated by a piecewise polynomial function and then evolved to the next time level according to the integral form of the conservation law. Godunov-type schemes are divided into two big classes: upwind and central ones. The evolution step of upwind schemes both the original first-order Godunov scheme [5] and its higher-order extensions (see, e.g., [4, 9,, 2]) are based on (approximate) solver of the (generalized) Riemann problem. Unfortunately, no Riemann problem solvers are available for the global flux model (.4), and therefore, it is unclear how upwind schemes can be applied to it. On the contrary, central schemes are based on the integration of conservation laws over the space-time control volumes that are selected so that each Riemann fan is entirely contained in its own control volume (this is achieved thanks to the finite speed of propagation). Thus, no (approximate) Riemann problem solver is needed, and therefore central schemes can be extended to problems for which the solution of the Riemann problem is unknown. The first-order Lax- Friedrichs (LxF) scheme [3, 0] is a prototype of central schemes. It is probably the most universal numerical method for time-dependent PDEs, but its resolution is quite low. The performance of the LxF scheme has been enhanced in [9], where the second-order central scheme the Nessyahu-Tadmor (NT) scheme was constructed by incorporating a second-order piecewise polynomial reconstruction into the LxF central framework. In this paper, we extend the (staggered) LxF and NT schemes to the conservation law with the global flux, (.4). The new schemes, derived in 2, are applied to a number of test problems in 3. In Example, we demonstrate robustness and high resolution, achieved by the proposed method. We also perform a numerical convergence test, comparison of the solutions of the equations with global and non-global fluxes, and dependence of the numerical solution on the look-ahead distance γ. Example 2 is designed to numerically study a dispersive effect of the global flux, which transforms a shock wave into a dispersive wave package. Finally, in Example 3, we numerically investigate a breakdown phenomenon and discover that when γ is small, the smoothing effect of the global flux seems to be able to prevent the shock discontinuity formation. 2 Derivation of the Central Schemes We first introduce the following notation: x α = x min + (α /2) x, t β = β t, where x and t are small spatial and temporal scales assumed, for simplicity, to be constants throughout the computational domain [x min, x max ], divided into the cells C j := [x j, x 2 j+ ] so that 2 J x max = x min + J x and [x min, x max ] = C j. Let us assume that at a certain time level t = t n, the solution, realized by its cell averages, ū n j := u(x, t n ) dx, x C j is available (we assume that the solution is compactly supported and that the computational domain [x min, x max ] is sufficiently large so that suppu(, t) [x min, x max ] for all t [0, T ], where T is the final time). We then construct its piecewise linear interpolant ũ n (x) := ū n j + sn j (x x j), x C j. (2.) j=

4 A. Kurganov and A. Polizzi This reconstruction is (formally) second-order provided the slopes s n j are (at least) first-order approximations of the derivatives u x (x j, t n ). A non-oscillatory nature of the interpolant (2.) is achieved by computing the slopes with the help of a nonlinear limiter. A library of such limiters is available (see, e.g., [4, 9,, 2, 3, 9, 22]), and one can compute the numerical derivatives using one s favorite limiter. In our numerical experiments, we have used the generalized minmod reconstruction [, 3, 9, 22] with: ( s n j = minmod θ ūn j ūn j, ūn j+ ūn j, θ ūn j+ ) ūn j, θ [, 2], (2.2) x 2 x x where the minmod function is defined as: min j {z j }, if z j > 0 j, minmod(z, z 2,...) := max j {z j }, if z j < 0 j, 0, otherwise, and the parameter θ can be used to control the amount of numerical viscosity present in the resulting scheme. It is well-known (see, e.g., [3]) that larger values of θ correspond to less dissipative but, in general, more oscillatory reconstructions. In our numerical experiments, we have used θ = 2. The reconstructed solution, ũ n (x), is then evolved to the next time level t = t n+ by integrating equation (.4) over the space-time control volumes [x j, x j+ ] [t n, t n+ ], which, after division by x, results in: ū n+ j+ 2 = x x j+ x j ũ n (x) dx x t n+ t n [ ] F (u(x j+, t), U(x j+, t)) F (u(x j, t), U(x j, t)) dt, (2.3) The first integral on the right-hand side (RHS) of (2.3) can be computed exactly. The flux integrals on the RHS of (2.3) should be computed using the (approximate) solution of the initial value problem (IVP) (.4),(2.) on the time interval t (t n, t n+ ) with the initial data prescribed at t = t n. Due to the finite speed of propagation, the solution of this IVP remains smooth at x = x j for all j provided the following CFL condition is satisfied: λa 2, t λ := x, where a is the largest local speed that can be estimated by: a V m max 0 u 2u = V m. Therefore, the flux integrals in (2.3) can be safely approximated by the mid-point quadrature leading to: ū n+ j+ 2 = ūj + ū j+ + x 2 8 (sn j sn j+ ) ] λ [F (u(x j+, t n+ 2 ), U(xj+, t n+ 2 )) F (u(xj, t n+ 2 ), U(xj, t n+ 2 )), (2.4)

Central Scheme for a Traffic Flow Model 5 where the values of both the solution u and its antiderivative U at the intermediate time level t = t n+ 2 can be approximated using the Taylor expansion, namely: u(x j, t n+ 2 ) ũ n (x j ) + t 2 u t(x j, t n ), U(x j, t n+ 2 ) Ũ n (x j ) + t 2 U t(x j, t n ). (2.5) In order to use these formulae, we need to provide the details on evaluating their RHSs. First, from equation (2.) we have: ũ n (x j ) = ū n j, and by integrating equation (2.) from x min to x we obtain the piecewise quadratic approximation of the antiderivative U: Ũ n (x) = x x min j = x j ũ n (ξ) dξ = x ū n i + i=0 i=0 x x j 2 ũ n (ξ) dξ ū n i + ūn j (x x j ) + sn j 2 2 (x x j 2 and, in particular, its point-values used in (2.5): )(x x j+ ), x C j (2.6) 2 U n j j := Ũ n (x j ) = x ū n i + x 2 ūn j ( x)2 s n j 8. (2.7) i=0 The time derivative u t in (2.5) is evaluated with the help of equation (.2): u t (x j, t n ) = F (u(x j, t n ), U(x j, t n )) x, (2.8) and the space derivative F x in (2.8) is computed using the minmod limiter: ( F (u(x j, t n ), U(x j, t n )) x = minmod θ F (ūn j, Uj n ) F (ū n j, Uj ) n, x F (ū n j+, Uj+) n F (ū n j, Uj ) n 2 x Finally, we integrate equation (2.8) with respect to x to obtain, θ F (ūn j+, U n j+) F (ū n j, U n j ) x ).(2.9) U t (x j, t n ) = F (u(x j, t n ), U(x j, t n )) = F (ū n j, U j n ), (2.0) needed to complete the evaluation of u and U at the intermediate time level in (2.5). Remarks.. Notice that a global approximation of U at time level t n, given by (2.6), allows one to easily compute the global flux F (ū n j, U n j ) = V m ū n j ( ū n j ) exp ( Ũn (x j + γ) Ũ n (x j ) γ ),

6 A. Kurganov and A. Polizzi even if x j + γ is not a grid node. 2. In practice, it is convenient to compute the point values of Uj n recursively, that is, replacing (2.7) with: Uj n = U j n + x 2 (u j + u n j ) + ( x)2 (s n j 8 sn j ). 3. The obtained second-order NT scheme for (.4) is given by (2.4),(2.) (2.2),(2.5) (2.0). It reduces to the first-order staggered LxF scheme if all the slopes are set to zero and the mid-point quadrature, used in the evaluation of the flux integrals in (2.3), is replaced with the left-sided rule. The resulting first-order scheme is: ū n+ j+ 2 = ūj + ū j+ 2 ] λ [F (u nj+, U nj+ ) F (unj, U nj ). (2.) 4. Since the derived schemes use alternating, staggered grids, one has to distinguish between the odd and even time steps. The formulae (2.4),(2.) (2.2),(2.5) (2.0) and (2.) describe the odd steps of the second- and first-order schemes, respectively. The even steps are obtained by shifting the indexes in the aforementioned equations by. Obviously, the computational 2 domain should be extended by x from both sides at every even step. This requires a careful 2 implementation of the given boundary conditions. In our numerical examples, the solution was constant at the edges of the computational domain so that we used these constant values at the computational cells near the boundary. 3 Numerical Examples In this section, we demonstrate a performance of the proposed central scheme on a number of test problems. In all numerical examples, the CFL number is 0.475, V m = 4, and the computational domain is taken large enough so that no waves are getting to its boundary within the final computational time. Example Red Light Traffic We consider equation (.4) with different γ s subject to the following initial data: u(x, 0) = {, 4 < x < 6, 0, otherwise. (3.) This corresponds to a situation in which the traffic light is located at x = 6 and it is turned from red to green at the initial time moment. We apply the proposed central schemes to the IVP (.4),(3.) and compute its solutions for different values of γ at time t =. We first fix γ = and study the behavior of our firstand second-order schemes. The solutions, computed on two uniform grids with x = /5 and x = /0, together with the reference solution, obtained by the second-order central scheme on a much finer uniform mesh with x = /400, are plotted in Figure 3.. The numerical convergence of the second-order scheme is demonstrated in Figure 3.2: the mesh is further refined to x = /20 and x = /40, and the solution seems to nicely converge to the reference solution.

Central Scheme for a Traffic Flow Model 7 x=/5 x=/0 0.6 ORDER 2 ORDER REFERENCE 0.6 ORDER 2 ORDER REFERENCE 0.4 0.4 0.2 0.2 0 4 6 8 0 0 4 6 8 0 Figure 3.: First- and second-order results on two different uniform grids. 0.6 x=/20 x=/40 x=/400 0.6 0.4 0.2 0 4 6 8 0 0.4 0.2 x=/20 x=/40 x=/400 0 3.9 4 4. 4.2 Figure 3.2: Second-order results on two finer uniform grids (left); zoom at the shock area (right). We then compare the computed solutions of the IVP (.4),(3.) and of a similar problem, but with non-global flux: (.),(3.). Both solutions are obtained by the second-order central scheme on the same uniform mesh with x = /20. The solutions are shown in Figure 3.3. One can clearly see the effect of the global flux that models the look-ahead dynamics. The effect is more prominent at the back end of the wave: the density wave corresponding to the global case clearly lags behind that of the non-global one. This can be explained by the awareness of traffic ahead. At the same time, since the interaction potentials vanish for vehicles at the front of the wave, the global and non-global solutions are almost identical there. We finally demonstrate the dependence of the computed solution on the look-ahead distance γ. The solutions, obtained by the second-order central scheme for γ = 0., 0.2, 0.5,, and on the same uniform grid with x = /40 are presented in Figure 3.4. As one can clearly see there, the dependence on γ is significant: as γ decreases, the drivers can see less, that is, their sensitivity to interaction is lowered and the original traffic jam dissipates more slowly. On the contrary, when

8 A. Kurganov and A. Polizzi γ is very large, the effect of the global flux is negligibly small. GLOBAL FLUX NON GLOBAL FLUX 0.6 0.4 0.2 0 4 6 8 0 Figure 3.3: An impact of the look-ahead dynamics: global vs. non-global fluxes. 0.6 γ=0. γ=0.2 γ=0.5 γ= 0.6 γ=0 γ=5 γ=2 γ= 0.4 0.4 0.2 0.2 0 4 6 8 0 0 4 6 8 0 Figure 3.4: Dependence of the solution on γ. Example 2 Traffic Jam on a Busy Freeway In the second example, we consider a different initial car density distribution: u(x, 0) = {, 6 < x < 8, 0.75, otherwise, (3.2) and study the behavior of the solutions of the IVP (.4),(3.2) for different γ s. Our numerical experiments clearly demonstrate convergence of the computed solutions as the grid is refined. We therefore omit the numerical convergence demonstration and all the presented results in this

Central Scheme for a Traffic Flow Model 9 example were obtained by the second-order central scheme on one particular uniform grid with x = /40 only (except for the last figure, where the smoothness of the solution is numerically investigated using finer grids). The dispersive effect, which could have been observed in Example, becomes much more prominent here since the waves are moving in the direction opposite to the vehicles velocity. Even in the case of a relatively large γ = 0, the solution of the IVP (.4),(3.2) (Figure 3.6) is significantly different from the solution of the corresponding IVP with a non-global flux (.),(3.2) (Figure 3.5). We then take a smaller γ = 5 and obtain quite similar solution, though the dispersive waves are now getting together by time t = 0, see Figure 3.7. We further decrease γ and discover a very interesting behavior of the solution. The results for γ = 3 are plotted in Figure 3.8 (note that the x-scaling is different in this figure compared with the one used in Figures 3.5 3.7). Even though the waves are still initially separated, the larger waves catch the smaller ones much faster now and a dispersive wave package is formed at about t = 3. When the look-ahead distance reduces to γ =, no separate waves are formed, but a dispersive wave package develops right away and is later transformed into one wave, see Figure 3.9 (notice that the x-scaling has been changed once again). When γ is reduced to 0.5, the dispersive effect can be observed at small times only (Figure 3.0), and for even smaller γ = 0., no dispersive effect can be observed (Figure 3.). In the latter case, the global flux leads to a certain smoothing effect though our fine mesh calculations suggest that the solution still contains a sub-shock at the left edge (front) of the wave, see Figure 3.2. t=0 t=6 t=3 t= Figure 3.5: Time evolution (from right to left) of the solution of (.),(3.2) with γ = 0. t=0 t=6 t=3 t= Figure 3.6: Time evolution (from right to left) of the solution of (.4),(3.2) with γ = 0.

0 A. Kurganov and A. Polizzi t=0 t=6 t=3 t= Figure 3.7: Same as in Figure 3.6 but with γ = 5. t=0 t=6 t=3 t= Figure 3.8: Same as in Figures 3.6 3.7 but with γ = 3. t=0 t=6 t=3 t= 5 0 5 0 5 5 0 5 0 5 5 0 5 0 5 5 0 5 0 5 Figure 3.9: Same as in Figures 3.6 3.8 but with γ =. t=0 t=6 t=3 t= 0 5 0 5 0 5 0 5 0 5 0 5 0 5 0 5 Figure 3.0: Same as in Figures 3.6 3.9 but with γ = 0.5.

Central Scheme for a Traffic Flow Model t=0 t=6 t=3 t= 0 5 0 5 0 5 0 5 0 5 0 5 0 5 0 5 Figure 3.: Same as in Figures 3.6 3.0 but with γ = 0.. x=/200 x=/400 5 5 5 5 0.75 0 2 4 6 0.75 0 2 4 6 Figure 3.2: Solutions of (.4),(3.2) with γ = 0. at t = 0 on two different fine grids. Example 3 Numerical Breakdown Study Our final example is devoted to the numerical investigation of the smoothing effect of the global flux. As it has been shown in Example 2, the global flux has an apparent dispersive effect for the large and intermediate values of the look-ahead parameter γ, while for small γ s a smoothing effect starts dominating. Though the rigorous analysis of the model (.4) is unavailable, the nature of the aforementioned phenomena can be heuristically explained by expanding U(x + γ) into the Taylor series about x and rewriting equation (.4) as follows: u t + [ V m (u u 2 )e TS] ( = 0, TS := u + γ ) x 2 u x + γ2 6 u xx +.... (3.3) Obviously, this can be done only for smooth solutions, for which (3.3) can be written in an equivalent nonconservative form: ( ) γ u t + V m ( 3u + u 2 )e TS u x = V m (u u 2 )e TS 2 u xx + γ2 6 u xxx +..., (3.4) which suggests that for sufficiently small γ, the (nonlinear) dispersive term on the RHS of (3.4) is small compared to the (nonlinear) viscosity term. This leads us to a conjecture that for small

2 A. Kurganov and A. Polizzi γ initially smooth solutions may remain smooth for all t. However, since in this case the viscosity coefficient is also small, smooth solutions may develop sharp gradients. To check our conjecture, we consider the following smooth initial data: u(x, 0) = 0.75 + 0.25e (x 7)2, (3.5) and use our second-order scheme to compute the solution of the IVP (.4),(3.5) on a uniform grid with x = /800. The final time is t = 0. In Figure 3.3 (left), we show the solution of this IVP for γ = 0. and. One can clearly see that the solution with γ = 0. remains smooth, while the solution with γ = seems to contain a shock discontinuity at the top part of the front (since the strength of the shock is very low, it is quite hard to capture it numerically, but a mesh refinement study, performed on two grids with x = /600 and x = /3200, suggests that there is a discontinuity there, see Figure 3.3 (right)). We compare these results with the solution computed with γ = 0 and with the solution of the non-global model (.),(3.5). As expected, both solutions, while being very different in nature, break down in finite time, see Figure 3.4. x=/800 γ= 5 γ=0. γ= 6 x=/600 x=/3200 4 5 2 0.75 2 0 2 4 6 8 2 3 4 5 Figure 3.3: Solutions of (.4),(3.5) with γ = 0. and γ = (left) and zoom at the finer mesh results for the γ = case (right). To further validate our conjecture of a strong smoothing effect of the global flux with small γ s, we demonstrate the time evolution of the computed solution of the IVP (.4),(3.5) with γ = 0., see Figure 3.5. As before, the grid is uniform with x = /800, and the solution does not seem to break down at any stage of its evolution. We have also performed finer grid computations (the results are not shown here) which fully support our conjecture. Acknowledgment: The work of A. Kurganov was supported in part by the NSF Grants # DMS-030585 and DMS-060430.

Central Scheme for a Traffic Flow Model 3 8 6 γ=0 NON GLOBAL FLUX 4 2 0.78 0.76 5 0 5 0 5 0 Figure 3.4: Solutions of (.4),(3.5) with γ = 0 and of (.),(3.5). t=20 t=4 t=8 t=2 Figure 3.5: Time evolution (from right to left) of the solution of (.4),(3.5) with γ = 0.. References [] M. Bando, K. Hasebe, A. Nakayama, A. Shibata, and Y. Sugiyama, Dynamical model of traffic congestion and numerical simulation, Phys. Rev. E, 5 (995), pp. 035 042. [2] M. Cremer and J. Ludwig, A fast simulation model for traffic flow on the basis of boolean operations, Math. Comput. Simulation, 28 (986), pp. 297 303. [3] K.O. Friedrichs, Symmetric hyperbolic linear differential equations, Comm. Pure Appl. Math., 7 (954), pp. 345 392. [4] E. Godlewski and P.-A. Raviart, Numerical approximation of hyperbolic systems of conservation laws, Springer-Verlag, New York, 996. [5] S.K. Godunov, A difference method for numerical calculation of discontinuous solutions of the equations of hydrodynamics, Mat. Sb., 47 (959), pp. 27 306. [6] D. Helbing, Gas-kinetic derivation of Navier-Stokes-like traffic equations, Phys. Rev. E, 53 (996), 2366 238.

4 A. Kurganov and A. Polizzi [7] D. Helbing and M. Treiber, Gas-kinetic-based traffic model explaining observed hysteretic phase transition, Phys. Rev. Lett., 8 (998), pp. 3042 3045. [8] A. Klar and R. Wegener, Kinetic derivation of macroscopic anticipation models for vehicular traffic, SIAM J. Appl. Math., 60 (2000), pp. 749 766. [9] D. Kröner, Numerical Schemes for Conservation Laws, Wiley, Chichester, 997. [0] P.D. Lax, Weak solutions of nonlinear hyperbolic equations and their numerical computation, Comm. Pure Appl. Math., 7 (954), pp. 59 93. [] B. van Leer, Towards the ultimate conservative difference scheme, V. A second order sequel to Godunov s method, J. Comput. Phys., 32 (979), pp. 0 36. [2] R. LeVeque, Finite volume methods for hyperbolic problems, Cambridge Texts in Applied Mathematics, Cambridge University Press, 2002. [3] K.-A. Lie and S. Noelle, On the artificial compression method for second-order nonoscillatory central difference schemes for systems of conservation laws, SIAM J. Sci. Comput., 24 (2003), pp. 57 74. [4] M.J. Lighthill and G.B. Whitham, On kinematic waves. II. A theory of traffic flow on long crowded roads, Proc. Roy. Soc. London. Ser. A, 229 (955), pp. 37 345. [5] M. Muramatsu and T. Nagatani, Soliton and kink jams in traffic flow with open boundaries, Phys. Rev. E, 60 (999), pp. 80 87. [6] K. Nagel and M. Paczuski, Emergent traffic jams, Phys. Rev. E, 5 (995), pp. 2909-298. [7] K. Nagel and M. Schreckenberg, A cellular automaton model for freeway traffic, J. Phys. I France, 2 (992), pp. 222 2229. [8] P. Nelson, Synchronized traffic flow from a modified Lighthill-Whitman model, Phys. Rev. E, 6 (2000), pp. R6052 R6055. [9] H. Nessyahu and E. Tadmor, Non-oscillatory central differencing for hyperbolic conservation laws, J. Comput. Phys., 87 (990), pp. 408 463. [20] G.F. Newell, Nonlinear effects in the dynamics of car flowing, Oper. Res., 9 (96), pp. 209 229. [2] P.I. Richards, Shock waves on the highway, Oper. Res., 4 (956), pp. 42 5. [22] P.K. Sweby, High resolution schemes using flux limiters for hyperbolic conservation laws, SIAM J. Numer. Anal., 2 (984), pp. 995 0. [23] A. Sopasakis, Formal asymptotic models of vehicular traffic. Model closures, SIAM J. Appl. Math., 63 (2003), pp. 56 584.

Central Scheme for a Traffic Flow Model 5 [24] A. Sopasakis, Stochastic noise approach to traffic flow modeling, Phys. A, 342 (2004), pp. 74 754. [25] A. Sopasakis and M.A. Katsoulakis, Stochastic modeling and simulation of traffic flow: asymmetric single exclusion process with Arrhenius look-ahead dynamics, SIAM J. Appl. Math., 66 (2006), pp. 92 944. [26] M. Treiber, A. Hennecke, and D. Helbing, Derivation, properties, and simulation of a gas-kinetic-based, nonlocal traffic model, Phys. Rev. E, 59 (999), pp. 239 253. [27] G.B. Whitham, Linear and Nonlinear Waves, Wiley, New York, 974.