TR No. 22 
for k = 20. Then we have 80% confidence that the correct value of the 
energy spectrum function is within the interval 
Ge. Ue ) 
Ba 2 Gig) 2 “enk) 
N42 ZA, 62. 
or that 
beh We) Bish SVAN 2 bee ZA. G8) epee 
The 80% confidence limits are indicated on the energy spectrum given 
in figure 23. The confidence limits for the other spectra are the same. 
Examination of the energy spectra indicates that the 80% confidence limits 
are reasonably correct. 
The predominant characterisitc of the spectra is the linear range (on a 
plot of log @/&) as a function of log K ) extending from wave numbers of 
0.01 cm7+ to 0,06 cm™+, At larger wave numbers the computed values of 
are subject to large error because of the relatively high noise level. Since 
any actual variations among the spectra are considered negligible with respect 
to statistical variations, a composite spectrum was formed from the individual 
spectra to determine more certainly the existence of the linear range: 
Keon (Ue) 
Kin (2) 685) 
=) Hb af / / 
Fn (i= 5) as (k ) ) Giese Cin 
p= | 
The composite spectrum is shown in figure 29. The effective sample length 
is 37 times longer than that of the individual samples, and the equivalent 
number of degrees of freedom is 740. The 80% confidence limits are indicated 
on the spectrum, Several of the individual spectra display secondary maxima 
at wave numbers ranging from 0.02 em~1 to 0.03 cm. This feature, however, 
is not apparent on the composite spectrum; so no significance is attached to 
it. 
If the approximate noise level, as estimated from the composite spectrum, 
is taken as SOO 
N61 ca ? a 
—e f ~ rd 
We (Golan ~ $6 Cm —See , 
eal 
and a noise correction applied to the composite spectrum, the result is as 
shown in figure 30. Within the range of wave numbers from #= 0.01 cm to 
[2 = 0.026 om™ >, the composite spectrum is of the expected form, viz: 
nw Se 
Gy (ke) OK 
24 
