## Thermostats by stochastic collision
### Andersen thermostat
- System is coupled to heat bath
- The coupling is represented by stochastic force. Impulsive force that act occasionally on random selected particles.
- Coupling strength described by collision frequency. And the collision follows Poisson distribution (time uncorrelated collisions) $P(t, \nu)=\nu e^{(-\nu t)}$ $\nu$ is frequency of collision.
>[!Note]
> The redistribution of velocity follows Maxwell Boltzmann distribution, so it is also not suitable for calculating kinetic related term (e.g. diffusion coefficient), except no collision happened.
Pros of Anderson thermostat:
- Generates canonical ensemble (NVT)
- Stochastic
- Local thermostat
Cons:
- Destroy momentum transport. No true molecular kinetics gets presented. Cannot be used to calculate transport properties.
### Langevin thermostat
- The system is sought to interact with the solvent (of small frictional particles), giving damping force.
- Pipping coefficient and violence of random force connected to fluctuation, dispersion relation, generate Canonical ensemble.
Pros:
- Recover NVT
- Local thermostat
- Ergodic
- Large time step possible
Cons:
- No transport properties
>[!Note]
> More suitable for simulating the reaction/process in solution.
### Nosé–Hoover thermostat
In general, this method introduces an additional term to control the behavior of thermal baths (effective mass & coupling), and made a conserved quantity, i.e., extended Hamiltonian.
#### Nosé–Hoover (2nd order on K):
- Proper sampling
- Deterministic (can be non-ergotic)
- Take another on K, can be oscillating
>[!Notice]
> The deterministic here is a very good property, meaning it can be not stochastic.
#### Nosé–Hoover chain (higher order on K):
- Canonical
- Ergodic
- Additional term for chaotic behavior