Fitting a local level Poisson (State Space Model)

I am working through "Forecasting with Exponential Smoothing". I am stuck on exercise 16.4 on the part that states:

The data set partx contains a history of monthly sales of an automobile part. Apply a local Poisson model. Parameters should be estimated by either maximizing the likelihood or minimizing the sum of squared errors.

The local Poisson model is defined as:

where and

I have the following code, but it seems to be stuck. The optimization always returns something close to the starting values.

Am I fitting the local Poisson model correctly?

library(expsmooth)
data("partx")
S <- function(x) {
  a <- x[1]
  if(a < 0 | a > 1)
    return(Inf)
  n <- length(partx)
  lambda <- numeric(n+1)
  error <- numeric(n)
  lambda[1] <- x[2]
  for(i in 1:n) {
    error[i] <- partx[i]-rpois(1,lambda[i])
    lambda[i+1] <-   (1-a)*lambda[i] + a*partx[i]
  }
  return(sum(error^2))
}

# returns a = 0.5153971 and lambda = 5.9282414
op1 <- optim(c(0.5,5L),S, control = list(trace = 1))
# returns a = 0.5999655 and lambda = 2.1000131
op2 <- optim(c(0.5,2L),S, control = list(trace = 1))