Monte Carlo
Algo for performing Monte Carlo:
- Initialization: Fix the number of MC steps, Choose initial \(R\) and variational parameters \(\alpha\) and calculate \(|\psi_T^\alpha (R)|^2\).
- Initialise the energy and the variance and start the MC calc.
- Calc a trail pos \(R_p = R + r \cdot step\) wehere \(r\) is a random var $r \in [0,1].
- Metropolis algo to accept or reject this move \(w = P(R_p)/P(R)\)
- If the step is accepted, then we set \(R = R_p\)
- Update averages
- Finish and compute final averages
General MCMC
- We have ProbDist (possibly high dimensionality). Want to sample from it or find expected value of it. Analytical solution is implausible.
- MCMC:
- Perform “random walk” through probdist favoring values with higher probs.
- Starting point -> pick random nearby point -> evaluate its probability. If it has higher prob than start: move there, otherwise stay (or move to that point with low prob)
- Then we are visiting every point in the probdist propotional to how probable that point is.
Markov chain
- Can be said in two ways:
- The next state is only dependent on the current state
- The current state is only dependent on the previous state
- Requirement:
- The chain has to be ergodic:
- It must visit every point in the domain, and will visit tem a proportionate amount to their probability.
- To be ergodic, the chain must be:
- Irreducible - for every state there is a positive probability of moving to any other state
- Aperiodic - the chain must not get trapped in cycles
- The chain has to be ergodic:
Variational Monte Carlo
- Can include functions from different sub-files
- Must have some initial variables: Seed, number of: particles, steps, dimensions
- How to choose step size(for Harmonic Oscillator):
- SE from the ??
- \(k(r) = u(r)/r\)
- \(r(0) = const\), \(r(\infty) = 0\)
- \(u(0) = 0\), \(u(\infty) = 0\)
- Harmonic Oscill: \(-\frac{\hbar^2}{2m} \frac{d^2}{dr^2}u(r) + 1/2 m \omega^2r^2u(r) = E u(r)\)
- Scale equations, make them dimensionless (to atomic units)
- \(\rho = \gamma r\) , \([r] = length\), \([\gamma] = length^{-1}\)
- \(r = \rho/\gamma\)
- \(-\frac{\hbar^2\gamma^2}{2m} \frac{d^2}{d\rho^2}u + 1/2 m \omega^2\rho^2/\gamma^2 u = E u\)
- Can choose and tune \(\gamma\) as we want to remove all the constants. Done by setting all constants = 1 and solving for gamma.
- Here: \(\gamma = \sqrt{\hbar/m\omega}\)