[R] non-stationary ar part in css

[Benedikt Gehr]

Hi

I would like to use arima () to find the best arima model for y time 
series. The default in arima apparently is to use conditional sum of 
squares to find the starting values and then ML (as described on the 
help page).
Now using the default may lead to error messages saying: "non-stationary 
ar part in CSS". When changeing the default to "ML" only the 
minimization works. As far as I understand, arima doesn't require 
stationarity, but apparently CSS does.
Can anyone tell me what exactly the css method does? And why is CSS-ML 
the default in R? Out of efficiency reasons? Because ML and ML-CSS gives 
the exact same estimates when applied to the same data. I tried to find 
out on google but I couldnt' find anything usefull or understandable to 
me as a non-statistician.

Here some data that causes the error message:

X<-6.841067, 6.978443, 6.984755, 7.007225, 7.161198, 7.169790, 7.251534, 
7.336429, 7.356600, 7.413271, 7.404165, 7.480869, 7.498686, 7.429809, 
7.302747, 7.168251,
7.124798, 7.094881, 7.119132, 7.049250, 6.961049, 7.013442, 6.915243, 
6.758036, 6.665078, 6.730523, 6.702005, 6.905522, 7.005191, 7.308986)

model.examp<-arima(X,order=c(7,0,0),include.mean=T)    # gives an error
model.examp<-arima(X,order=c(7,0,0),include.mean=T,method="ML")  # gives 
no error

Any help on this would be most appreciated

Many thanks fo the help

best wishes

Benedikt