New Traffic Model Tames Wild Speed Swings by Letting Gaps Drive the Math

Traffic models that promise to describe how congestion forms and dissolves have long carried a hidden flaw: they assume every driver reacts instantly and in exactly the same way. The new formulation, called the Benjamin-Aaron-Mayumbo, or BAM, model, extends the classical Payne-Whitham framework and, in simulations on a closed-loop roadway, kept predicted speeds within physically plausible bounds where the classical model produced impossible values. The work, published as an open-access research article on 8 October 2026 in Discover Networks, was carried out by Benjamin Simwinga, Aaron Zimba and Mayumbo Nyirenda.

The starting point is the Payne-Whitham model, a second-order macroscopic traffic flow formulation that treats traffic as a continuous medium described by density and mean speed. Macroscopic models of this kind are widely used because they represent collective traffic behaviour computationally efficiently, in contrast to microscopic models that simulate individual vehicles and mesoscopic approaches that blend the two. The Payne-Whitham model adds a velocity equation to the conservation of vehicles, containing convective acceleration, a traffic-pressure term that captures the effect of vehicle interactions, an equilibrium speed-density relationship and a relaxation time describing how quickly drivers adjust toward the equilibrium speed. That relaxation time, however, is treated as a fixed parameter, and driver response is assumed uniform across vehicles, an assumption the authors note can yield unrealistic predictions in heterogeneous traffic.

The BAM model replaces that fixed relaxation structure with a gap-controlled one. The forward distance headway, the spacing a vehicle maintains from the vehicle ahead in the same lane, is approximated at the macroscopic level by the inverse of traffic density, an approximation the authors state assumes relatively homogeneous vehicle spacing. From this spacing, the model derives a gap-controlled desired velocity, in which the attainable speed rises toward the free-flow maximum as the gap grows and falls as the gap shrinks. At the same time, an effective relaxation time is introduced that varies with density and available spacing, so that the rate at which drivers adjust their speed also responds to traffic conditions. The two relationships combine into a single density-dependent acceleration term that replaces the classical relaxation term in the Payne-Whitham momentum equation, while the original traffic-pressure mechanism is retained. The authors are careful to note that the effective relaxation time is a state-dependent adjustment timescale, not an explicit perception-reaction delay, since no time-delayed terms appear in the formulation.

Because the new model modifies only the source term while leaving the flux structure untouched, its mathematical character closely mirrors the original. The Jacobian matrix of the flux vector, and therefore the characteristic wave speeds, are identical for both models, and both share real, distinct eigenvalues that preserve strict hyperbolicity. This allowed the researchers to apply the Roe approximate Riemann solver, a finite-volume method commonly used for hyperbolic conservation laws and traffic flow because of its ability to capture discontinuities and shock waves, uniformly to both formulations. Harten’s entropy fix was applied to prevent unphysical expansion shocks at sonic points where eigenvalues vanish, ensuring numerical stability and smooth dispersion of traffic fields downstream of bottlenecks.

Beyond numerics, the team derived an analytical linear stability condition for the new model. They considered a steady, homogeneous traffic stream perturbed by small localized disturbances and linearised the governing equations, obtaining a dispersion relation for monochromatic wave solutions. Expanding for long-wavelength disturbances yielded a kinematic wave propagation speed and a growth-or-decay parameter, from which the stability condition follows: the anticipation coefficient must be at least as large as a density-dependent threshold involving the safe spacing parameter, the free-flow speed and the local density. Because the threshold’s denominator is squared, it decreases sharply as traffic enters heavily congested zones, and because the effective relaxation time shortens at higher density, the model compresses the instability envelope compared with the classical Payne-Whitham model, whose fixed Greenshields-based stimulus produces a wide parabolic instability region where stop-and-go waves easily amplify at mid-to-high densities.

To check that this theoretical boundary actually holds, the researchers ran numerical experiments with parameters placed deliberately on both sides of it. For a mean density of 0.30 vehicles per metre, a free-flow velocity of 20 metres per second and a safe spacing of 8 metres, the critical anticipation coefficient came out at approximately 4.15 metres per second. Three simulation runs with identical initial sinusoidal perturbations but different anticipation coefficients confirmed the prediction: at 2.0 metres per second the density perturbation amplified into backward-propagating shock waves, at 20.0 metres per second the disturbance decayed exponentially, and at the critical value of 4.15 metres per second the perturbation amplitude remained largely constant, a neutrally stable case. The numerical findings, the authors report, directly align with the analytical prediction.

The headline simulation comparison was conducted on a 500-metre closed-loop roadway containing 50 identical vehicles with an inactive bottleneck, discretised into 500 one-metre cells with a time step of 0.006 seconds over a 6-second horizon, using a Greenshields equilibrium velocity-density relationship and a maximum velocity of 20 metres per second. The results were stark. Under the tested conditions, the classical Payne-Whitham model produced severe oscillatory behaviour, with unphysical speeds surging as high as 320 metres per second and dropping to minus 180 metres per second, values that are physically impossible. The BAM model, by contrast, maintained velocities within the prescribed range of 0 to 20 metres per second and exhibited substantially smoother velocity evolution. The authors stress that this comparison demonstrates more bounded and physically reasonable velocity behaviour under the tested simulation conditions, rather than establishing superior empirical accuracy solely from the observed velocity range.

Longer 50-second simulations of density evolution reinforced the contrast. Starting from a smooth sinusoidal density perturbation, the proposed model kept the density field relatively smooth throughout, with a local maximum of about 0.45 vehicles per metre near 150 metres and a minimum of about 0.15 vehicles per metre near 375 metres, indicating stable wave propagation and limited amplification of local disturbances. The classical model, under the same perturbation, progressively developed sharper variations, with a maximum density of roughly 0.42 vehicles per metre around 150 metres and a minimum near 0.17 vehicles per metre around 350 metres, steepening into shock-like structures and backward-propagating stop-and-go waves. Flow profiles told a similar story: both models produced comparable peak flows of about 4.55 vehicles per second and minima near 3.4 vehicles per second, but the PW profile lost its symmetry and deformed over time, while the proposed model preserved a smooth, harmonic structure. The authors note that the comparable flow magnitudes indicate the principal difference lies in the stability of traffic-state evolution rather than an increase in flow.

The team also probed robustness across four scenarios on roadways of 250, 500 and 1000 metres, varying density, perturbation magnitude, reaction timescale, standstill spacing and bottleneck conditions, and tracking the final-to-initial RMS density perturbation ratio as a damping indicator. Under baseline conditions the ratios were 0.002, 0.213 and 0.681 across the three lengths, showing stronger attenuation on shorter roadways. Raising the density and perturbation magnitude kept solutions bounded with similar ratios, while a longer reaction timescale and larger standstill spacing produced stronger damping, with ratios of roughly 0.000001, 0.051 and 0.487. An active bottleneck, modelled as a 25 percent local speed reduction over the central 10 percent of the road, produced the largest residual disturbance, with RMS ratios rising to 0.717, 1.332 and 1.923, indicating sustained spatial redistribution and amplification that grew with road length, though velocities remained bounded throughout.

The authors are explicit about the limits of these findings. The study lacks empirical calibration and validation, and the results, drawn from controlled numerical experiments with prescribed initial perturbations, demonstrate the mathematical and numerical behaviour of the formulation rather than its predictive accuracy under real traffic conditions. The gap-to-density transformation assumes homogeneous spacing and does not explicitly resolve differences in vehicle types, lengths or individual driver characteristics, and some parameter choices, such as a reduced reaction time of 0.052 seconds, were sensitivity parameters rather than behavioural estimates. Future work, the researchers write, will focus on calibrating the BAM parameters against observed traffic data and validating the model against independent measurements of flow, speed, density, headway and travel time, together with heterogeneous traffic conditions and more realistic network configurations. Until then, the model stands as a numerically disciplined extension of a classical framework, one whose promise will be tested not on a computer but on real roads.

Subject of Research: A gap-controlled macroscopic traffic flow model extending the Payne-Whitham framework with adaptive relaxation dynamics

Article Title: A macroscopic traffic flow model with gap-controlled driver response and adaptive reaction dynamics

Article References: Simwinga, B., Zimba, A., & Nyirenda, M. (2026). A macroscopic traffic flow model with gap-controlled driver response and adaptive reaction dynamics. Discover Networks, 2(1), Article 39. https://doi.org/10.1007/s44354-026-00053-y

Image Credits: AI Generated

DOI: 10.1007/s44354-026-00053-y

Keywords: traffic flow modelling, Payne-Whitham model, macroscopic traffic model, driver response, relaxation time, linear stability analysis, Roe approximate Riemann solver, stop-and-go waves, traffic simulation, headway, numerical stability, traffic congestion