A modeling & simulation ladder

Ten rungs, from a grid to a digital twin

Ten simulations, each in a new field and each inheriting every tool below it. It starts with a forest fire on a NumPy grid and ends with a 3D campus model driven by real bus GPS, real buildings, a real solar year and 17,579 real crystal structures. No rung counts as done until it has matched a known answer, from theory or published data.

Tools in the stack, cumulative
L012L024L037L0410L0512L0615L0719L0823L0927L1032
Each bar counts the libraries and methods a level uses, including everything it inherits.
  1. Wildfire Percolation: hero image
    Level 1 · Statistical physics · ecology

    Wildfire Percolation

    Cellular automata · Monte Carlo · phase transitions

    Why a forest at 59% tree density burns end to end and one at 55% does not. Plus self-organised criticality, where fires follow a power law nobody wrote into the rules.

    + NumPy+ Matplotlib
    VERIFIEDPercolation threshold 0.5936 ± 0.0007 vs 0.5927 in theory; forest-fire density 0.3978 vs 0.3976 (Grassberger 2002).
  2. Noise in Gene Circuits: hero image
    Level 2 · Systems & synthetic biology

    Noise in Gene Circuits

    ODEs vs exact stochastic simulation

    A genetic toggle switch that the deterministic equations say can never flip, which flips anyway once you count molecules one at a time.

    + SciPy solve_ivp+ Gillespie SSA
    VERIFIEDGillespie runs match the exact master equation (total-variation distance 0.007); switching time grows like e^(0.53 Ω).
  3. ATL Checkpoint: hero image
    Level 3 · Operations research

    ATL Checkpoint

    Discrete-event simulation · output analysis · optimisation

    A TSA checkpoint at the world's busiest airport: time-varying arrivals, PreCheck and secondary screening. What is the cheapest staffing that keeps the 95th-percentile wait under 15 minutes?

    + SimPy+ confidence intervals+ Erlang C
    VERIFIEDM/M/5 queue: Erlang C gives a 0.4789 min wait, the simulation 0.4793 ± 0.033. The robust plan needs 141 lane-hours vs 148 for textbook staffing.
  4. Echo Chambers: hero image
    Level 4 · Computational social science

    Echo Chambers

    Agent-based modelling on adaptive networks

    Bounded-confidence agents on a social network, plus a recommender that rewires ties toward agreement. When does a society reach consensus, polarise or shatter?

    + Mesa 3+ NetworkX+ parallel sweeps
    VERIFIEDOpinion-cluster counts follow ⌊1/2ε⌋ (Deffuant 2000) at all five ε tested. With the recommender, camps seal into disconnected networks in 100% of runs.
  5. Galactic Collision: hero image
    Level 5 · Astrophysics

    Galactic Collision

    N-body · symplectic integration · high-performance Python

    Two disk galaxies in live dark-matter halos swing past each other and throw out tidal tails, rendered to video with 16,000 bodies.

    + Numba JIT + prange+ ffmpeg video
    VERIFIEDKepler period error 0.025%. Leapfrog energy stays bounded while Euler drifts to 80%. Code runs 1,558× faster than plain Python.
  6. Virtual Wind Tunnel: hero image
    Level 6 · Fluid dynamics

    Virtual Wind Tunnel

    Lattice Boltzmann CFD · verification & validation

    A from-scratch lattice Boltzmann solver: Kármán vortex streets, a stalled airfoil and a lift/drag polar.

    + stencil kernels+ scipy.fft+ V&V
    VERIFIEDPoiseuille flow converges at exactly order 2.000. Schäfer–Turek Strouhal 0.3014 (range 0.295–0.305); cylinder Strouhal 0.1651 vs Williamson's 0.164.
  7. Earth's Thermostat: hero image
    Level 7 · Climate science

    Earth's Thermostat

    Bayesian calibration · uncertainty · global sensitivity

    A two-layer energy-balance model calibrated by MCMC on 146 years of NASA temperatures, projecting five IPCC futures with honest error bands.

    + pandas+ live data APIs+ emcee MCMC+ SALib Sobol
    VERIFIEDTransient climate response 1.60 [1.42, 1.80] K, inside the IPCC AR6 range. Recovers 10/10 synthetic truths; 93% of observations fall in the 90% band.
  8. Generative Bracket: hero image
    Level 8 · Structural engineering

    Generative Bracket

    Finite elements · 3D topology optimisation

    The GE jet-engine bracket challenge: a computer grows the stiffest titanium bracket for four load cases, torsion included, and exports it as a printable STL.

    + scikit-fem+ pyamg+ PyVista / VTK+ STL
    VERIFIEDCantilever tip deflection within 1% of Euler–Bernoulli; exactly 8.00% volume; the STL is watertight (351 g of Ti-6Al-4V).
  9. Orbital Traffic Control: hero image
    Level 9 · Astrodynamics · space safety

    Orbital Traffic Control

    Real catalogue · conjunction screening · debris Monte Carlo

    All 18,677 tracked objects on a live WebGL globe, the closest approaches over the next 24 hours, and a 200-year Kessler-syndrome outlook.

    + sgp4+ KD-trees+ Three.js+ satellite.js
    VERIFIEDTwo independent orbit codes (python-sgp4, satellite.js) agree to 0.25 m. Top conjunction: a 75 m miss at 15.1 km/s.
  10. Solar Stinger: hero image
    Level 10 · Energy · mobility · materials

    Solar Stinger

    Real-data multi-physics digital twin

    Real campus bus GPS, OpenStreetMap buildings, an NSRDB solar year and the Materials Project, coupled in one 3D twin. One roof could power the whole electric fleet, but only if the buses charge while the sun is up.

    + pvlib+ OpenStreetMap+ Materials Project+ NREL HTEM+ digital twin
    VERIFIEDHorizon ray-march 26.4° vs 26.3° analytic; SQ limit 33.69% at 1.34 eV vs 33.7%; solar yield 1,411 kWh/kWp vs PVWatts ~1,400.