a Stationary time series always introduces errors of a statis- 
tical nature. The techniques given by Tukey [1949] tell us how 
big the error is and how to make it smaller if desired. 
In addition, Tukey and Hamming [1949] discuss design criteria 
of physical wave analyzers. The numerical methods can be used 
to calibrate the wave analyzers, to tell how much in error the 
physical analysis is, and to determine possible improvements in 
the design of the instruments. More will be discussed on these 
points later. 
Statement of problem 
A wave record is a short section of a very nearly tempor- 
arily homogeneous lonzer record. It can to a first approximation 
be treated as a stationary process. Such a wave record could be, 
Say, seven minutes long (the usual Beach Hrosion Board practice), 
twenty minutes long (Barber and Ursell [1948]) or an hour long. 
Consider such a record. Read off the values of the record at, 
say, one second intervals of the record and tabulate the values in 
terms of their departure from the mean of the record at the time 
of observation. The result is a sample from a stationary time ser- 
ies, and there are N points representing N values as Eigen in equa- 
tion (10.25). The problem of numerical wave record analysis is 
to find an estimate of the function [ace )]° from these N numbers, 
and to tell how reliable this estimate is. 
OO ee ee ee eee EE ee EE Eee ee 
function 
The problem is to find LAC ele It will be found first by 
abstract methods, and then by practical numerical methods. The 
- 271 - 
