📘 How do physicists choose a model?

What you’ll learn

1. Continuum versus discrete modeling decisionsEvaluate trade-offs between continuum and discrete representations for a given physical scale and observable.Continuum fields smooth fluctuations while discrete lattices capture microscopic correlations exactly. The crossover is governed by the ratio of microscopic length to macroscopic gradient scale. Quantitative error bounds follow from renormalization-group analysis of the relevant operators.
2. Lagrangian and Hamiltonian structures in field theoryDerive conserved quantities from both Lagrangian and Hamiltonian field-theoretic descriptions.Symmetries of the action produce conserved currents via Noether’s theorem. The Hamiltonian density generates time evolution through Poisson brackets on the infinite-dimensional phase space. Canonical transformations preserve the symplectic structure while simplifying the equations of motion.
3. Boundary-value problems for elliptic PDEsClassify well-posedness of elliptic boundary-value problems arising in electrostatics and elasticity.Elliptic operators are Fredholm between appropriate Sobolev spaces. Boundary conditions determine whether the problem is self-adjoint or requires additional compatibility conditions. Green’s functions encode both existence and regularity of solutions.
4. Finite-element discretization of linear elasticityImplement and analyze finite-element schemes for static linear elasticity with mixed boundary conditions.Weak forms of the equilibrium equations are discretized on polynomial spaces satisfying the inf-sup condition. A priori error estimates depend on the regularity of the solution and the approximation properties of the element family. Locking is avoided by reduced integration or mixed formulations.
5. Monte Carlo sampling in statistical mechanicsDesign and diagnose Monte Carlo algorithms for equilibrium sampling of interacting particle systems.Detailed balance ensures the correct Boltzmann distribution is reached. Autocorrelation times quantify the efficiency loss from critical slowing down. Cluster and worm algorithms reduce these times from power-law to logarithmic scaling near continuous transitions.
6. Molecular dynamics integrators and thermostatsSelect and validate symplectic integrators together with thermostat or barostat schemes for production runs.Symplectic integrators preserve phase-space volume and yield bounded energy fluctuations. Thermostats enforce the correct ensemble by coupling to auxiliary variables whose equations derive from an extended Hamiltonian. Multiple-time-step methods separate fast and slow forces to reach microsecond scales.
7. Bayesian uncertainty quantificationConstruct Bayesian posteriors over model parameters given both experimental data and emulator uncertainty.Priors encode domain knowledge while likelihoods incorporate measurement covariances. Markov-chain Monte Carlo or nested sampling explores the posterior. Emulator discrepancy is treated as an additional source of uncertainty that widens credible intervals.
8. Proper orthogonal decomposition for model reductionBuild and assess POD-Galerkin reduced-order models for high-dimensional nonlinear PDEs.Snapshot matrices yield orthogonal modes ranked by captured energy. Galerkin projection produces a low-dimensional dynamical system whose stability depends on the neglected dissipative scales. Stabilization via closure models or Petrov-Galerkin weighting restores accuracy.
9. Multiscale coupling of atomistic and continuum domainsFormulate and stabilize concurrent multiscale simulations across atomistic-continuum interfaces.Handshake regions transmit conserved quantities while minimizing spurious reflections. Iterative force balancing or dynamic relaxation eliminates ghost forces. Error estimates track the mismatch between the atomistic stress tensor and the continuum constitutive law.
10. Model validation against high-precision experimentsDesign validation protocols that separate model inadequacy from experimental systematic error.Validation requires quantitative comparison of model predictions with data whose covariance is fully characterized. Discrepancy metrics such as the Mahalanobis distance isolate regions where the model fails. Iterative refinement updates both parameters and model structure.
11. Global sensitivity analysis via Sobol indicesCompute and interpret variance-based sensitivity measures for expensive physics simulators.Sobol indices decompose output variance into main effects and interactions. Surrogate models or polynomial chaos expansions reduce the number of required simulator evaluations. Indices greater than 0.05 typically identify parameters that merit further experimental constraint.
12. Subgrid-scale modeling in large-eddy simulationAssess the accuracy and stability of common subgrid-scale closures for incompressible turbulence.Filtering the Navier-Stokes equations produces an unclosed stress tensor whose divergence must be modeled. Dynamic procedures compute coefficients from scale similarity at the grid cutoff. Structural models incorporate additional physics such as scale-similarity or approximate deconvolution.
13. Many-body perturbation theory for electronic structureImplement and benchmark GW and Bethe-Salpeter calculations for quasiparticle and optical spectra.The self-energy operator in GW captures screened exchange and correlation beyond density-functional theory. The Bethe-Salpeter equation adds electron-hole attraction for optical response. Convergence with respect to k-point sampling and empty bands must be demonstrated to 0.01 eV.
14. Physics-informed neural networks for inverse problemsFormulate and train physics-informed networks for parameter inference in partial differential equations.The loss functional penalizes both data misfit and the PDE residual evaluated by automatic differentiation. Hard constraints on boundary conditions are imposed through network architecture. Training dynamics are governed by the relative weighting of data and physics terms.

Questions this course answers

Grounded in trusted sources

• OpenStax
• American Physical Society
• National Institute of Standards and Technology
• J. N. Reddy, An Introduction to the Finite Element Method
• Frenkel & Smit, Understanding Molecular Simulation
• Saltelli et al., Global Sensitivity Analysis — Sobol indices
• Raissi et al., physics-informed neural networks literature
• NASA / DOE verification & validation guides for computational models

Every Wunder lesson is built from real, reputable sources — never invented.

Related courses