Donsker’s invariance principle

Donsker’s invariance principle is a classical extension of the central limit theorem. It says that the path of a discrete random walk converges, if properly rescaled, in distribution to a standard Brownian motion. More details can be found in your favorite text book or on wikipedia. Here just comes an animation.

library(dplyr)
library(gganimate)
library(transformr)
set.seed(42)
n <- 10000  # maximal steps of random walk  
n_seq <- seq(from=10, 100, by = 10)  # starting phase
n_seq <- c(n_seq, seq(from=100, n, by = 100))
number_iteration <-length(n_seq)   
X <- sample(c(-1,1), n, replace=TRUE, prob=c(0.5, 0.5))
S <- cumsum(X)  # random walk S_n

W <- numeric(number_iteration*n)   
id <- numeric(number_iteration*n)      
time <- numeric(number_iteration*n)   
for (i in 1:number_iteration){
  for (j in (1:n)){
    index <- floor(j/n * n_seq[i]) + 1
    W[j+(i-1)*n] <- (S[index] + ((j/n * n_seq[i]) -(index-1)) *( S[index+1]-S[index]))/ sqrt(n_seq[i])
  }
  id[(1+n*(i-1)):(n*i)] <- rep(n_seq[i],n)
  time[(1+n*(i-1)):(n*i)] <- (1:n)/n
}

rw <- data.frame(x = time,
                     y = W,
                     id =id)

p <- ggplot(rw, aes(x=x, y=y)) + 
  geom_line()+ 
  ylab(expression(W^{(n)}~ (x)))+
  transition_states(
    id,
    transition_length = 2,
    state_length = 1) +
  enter_fade() + 
  exit_shrink() +
  ease_aes('sine-in-out')+
  labs(title = "Donsker's invariance principle. n = {closest_state}") 

animate(p, nframes=number_iteration*2)