Reading path
City models: from forecasts to decisions
Start with a classroom timetable, then examine missing traffic states, numerical event errors and the limits of model reduction.
Three core readings: about 55 min · With the optional extension: about 77 min
Understand the problem, then examine the evidence.
The first three articles form the core. The fourth asks for more mathematical background and is optional. Reading times are estimates; there is no need to finish in one sitting.
Core 1 of 3About 9 min
When the Timetable Lies to the Ventilation System
See how room changes and shared fan limits challenge a forecast, and what an occupancy buffer costs.
- Useful starting points
- Proportions, graphs and a simple balance of what enters and leaves a room.
- Evidence and limits
- Results come from the study’s own synthetic generator. Fan energy is a proxy and independent field validation is still needed.
Core 2 of 3About 26 min
What Must a Traffic Model Remember?
Separate omitted state information, projection-induced memory and noise when a longer history improves predictions.
- Useful starting points
- Functions and time-series graphs; linear algebra and Fourier modes support the derivations.
- Evidence and limits
- The main comparison retains both headway and velocity modes in a synthetic 24-car ring. Fitted histories are not certified exact memory kernels.
Core 3 of 3About 20 min
When a Traffic Solver Invents a Jam
Check whether a small final global error also supports a threshold arrival time or a queue measure.
- Useful starting points
- Conservation and moving fronts. The finite-volume and differential-equation details are a deeper layer.
- Evidence and limits
- This dimensionless LWR benchmark retains a failed predeclared criterion. It is neither real traffic evidence nor a universal ranking of solvers.
Optional advanced readingAbout 22 min
When a Reduced Ventilation Model Leaves Its Training Regime
Ask whether a reduced model passes the nominal control before attributing errors to a change of regime.
- Useful starting points
- Matrices and approximation error; POD and DEIM are advanced details.
- Evidence and limits
- Both reduced models already fail the nominal holdout, so shifted trajectories remain diagnostics. This is not a CO₂ or infection-risk validation.