Why Peak‑Travel “Off‑Peak” Strategies Fail: Traffic Flow Theory Explained

Traffic‑flow fundamentals

Three macroscopic variables describe vehicular traffic:

Flow (q, vehicles / hour): number of vehicles passing a cross‑section per unit time.

Density (ρ, vehicles / kilometre): number of vehicles per unit length of road.

Speed (v, kilometres / hour): average speed of the vehicles.

These variables obey the conservation relationship q = ρ·v. When density is low, flow can be high; once density exceeds a critical value, speed drops sharply, flow declines, and a jam forms.

LWR model and backward‑propagating shock waves

The Lighthill‑Whitham‑Richards (LWR) model captures the spatio‑temporal evolution of density with a first‑order hyperbolic PDE: ∂ρ/∂t + ∂q(ρ)/∂x = 0 Using the Greenshields linear speed‑density assumption, v(ρ) = v_f \left(1 - \frac{ρ}{ρ_j}\right) where v_f is the free‑flow speed and ρ_j the jam density (maximum possible density). The resulting flow‑density (fundamental) diagram is a downward‑opening parabola:

q(ρ) = ρ·v_f \left(1 - \frac{ρ}{ρ_j}\right)

Maximum flow occurs at the critical density ρ* = ρ_j / 2, giving the road capacity q_{max} = \frac{v_f·ρ_j}{4} If the actual demand q_d exceeds q_{max}, a density shock forms. The shock speed is given by the Rankine‑Hugoniot condition: s = \frac{q_2 - q_1}{ρ_2 - ρ_1} Because q_2 < q_1 in congested conditions, s is negative, meaning the jam propagates upstream (against traffic flow). This explains why a blockage can appear far behind the actual point of overload.

Game‑theoretic view of off‑peak “peak‑shaving”

Assume N drivers choose a departure time t_i within a day. Their perceived travel cost can be modelled as C_i = α·Δt(ρ(t_i)) + β·|t_i - t^*| where Δt is the delay caused by the density at t_i, t^* is the desired arrival time, and α, β weight delay versus schedule deviation. Under homogeneous drivers, perfect information, and free choice of t_i, the Nash equilibrium equalises marginal cost across all time slots. Consequently, when many drivers shift to the night‑time “off‑peak” slot, its density rises, eroding the advantage and creating a new peak.

Rough capacity estimation for Chinese highways

A typical four‑lane expressway lane can handle about 1,800 – 2,000 veh/h. Using the midpoint 1,900 veh/h gives a total capacity of roughly 7,600 veh/h for the whole road. Over 24 hours this corresponds to about 180,000 veh per lane‑day.

The 2026 Spring Festival travel forecast reported a peak daily highway flow of ~3.5 million vehicle‑trips nationwide. Although the measurement scopes differ, the figure shows that during the return‑travel peak the instantaneous demand can approach the capacity of individual road segments.

Key conclusions

Congestion does not require an accident. Once density exceeds the critical threshold, a shock wave forms and travels upstream, creating a jam on a clear road.

Off‑peak advice loses effectiveness when it becomes common knowledge. Real‑time navigation apps and mass‑media dissemination turn the “optimal” departure time into a public signal, causing many drivers to converge on the same slot and generate a new peak.

Total travel volume is the fundamental driver. The 2026 Spring Festival generated an estimated 9.5 billion person‑trips, a ~10 % increase over the previous year. No individual timing strategy can offset such scale.

Free‑highway policies amplify demand. Removing tolls lowers monetary cost, attracting additional drivers and raising the peak flow.

Potential mitigation strategies

Individual‑level actions

Depart earlier than the widely advertised “night‑off‑peak” window, e.g., before the third day of the lunar new year, to stay ahead of the collective shift.

Prefer high‑speed rail for long distances; its schedule is fixed and immune to road congestion.

If driving is unavoidable, monitor traffic conditions for the next 100 km and be prepared to divert to national roads before entering a congested segment.

System‑level measures

Increase physical capacity. Widen lanes, redesign toll plazas, and promote electronic toll collection (ETC). Capacity gains diminish during extreme holiday peaks and involve high capital cost.

Flatten demand distribution. Congestion pricing—charging higher fees during peak periods—shifts elastic demand to off‑peak times, counteracting the effect of “free‑holiday” policies.

Accelerate automated driving. Higher automation can reduce headways and reaction delays, raising effective capacity, but significant impact requires a high market penetration of automated vehicles.

Broader context

The Spring Festival congestion reflects a structural mismatch between population distribution and economic activity: millions work in coastal megacities while families remain in inland provinces, producing massive seasonal migration. Long‑term solutions therefore involve balanced regional development, expanded public‑transport networks, and finer‑grained demand management, beyond merely widening roads.

Understanding the LWR model, the shock‑wave mechanism, and the game‑theoretic equilibrium clarifies why “off‑peak” strategies can quickly become ineffective, even though they may still offer modest personal benefits when information is imperfect.

References: Lighthill & Whitham (1955), Richards (1956), 2026 Spring Festival Travel Forecast, relevant Zhihu discussions (Feb 23 2026).