the difference between the forecasted value and the actual value 
squared is a minimum over all forecasts. If the autocorrelation 
function is essentially zero from lags of three minutes onward, then 
the forecast would be a zero amplitude disturbance at all times be- 
yond three minutes in the future. This forecast would be correct 
in the least squares sense because the second moment about the mean 
(zero) is the smallest second moment possible and in the sense that 
the autocorrelation function implies that what will happen in three 
minutes has nothing to do with what is happening. 
If it ever becomes essential to know thirty seconds in advance 
that a big wave is coming then it is possible to imagine an elect- 
ronic circuit constructed along the lines of the one described by 
Lee [1949] which will graph the wave record as it will occur 30 
seconds in the future given the present wave record. Note also 
figure 22 in Lee's paper. The random voltages shown look exactly 
like wave records!! The machine described by Lee [1949], if one 
imagines it applied to wave forecasts would only predict the records 
about three seconds in advance. 
A ship at sea is acted upon by a Gaussian wave system. There- 
fore it pitches, rolls, and rises and falls according to a Gaussian 
law. The continuous record of, say, the inclinometer is therefore a 
temporarily homogeneous Gaussian record, and from the autocorrelation 
function of the inclinometer record it is therefore possible to pre- 
dict from the past when the next big roll of the ship will occur. 
A very fruitful line of future research will be to apply the 
methods given by John [1949] to a Gaussian sea surface and determine 
the movement of floating objects on the sea surface in response to 
21 
