Thus the recursion for computing the cumulants is closed, and 
the indices A,P may be dropped if (9) is substituted in (11) 
and )10) in (12): 
  
  
Ky_q + pm 
A, = 34.4 T 3 ^ (y, “- 8,4) (13) 
k, 4 + Nt b^T 
; ; 2 
No (K, 4 + bÉT) 
K, = (14) 
N, + Ky_q+ DT 
From the recursion equation for the second cumulant (14) it 
is apparent. That K. must approach a stationary value, because no 
process observations are treated. The stationary value of the 
cunulants is determined only by estimations of the standard 
deviations of system and observations disturbances. If the 
iterative computation of the second cumulant is performed in 
the cyclic computation of the filter, the behaviour of the 
cumulant approaching its stationary value may be utilized 
for accelerating the minimization of errors that may occur 
because of faulty initial values for the first cumulant of 
the state variables. 
In order to make the filtering more lucid, let us intraduce 
the abbreviated notations 13% = K, and 
y= ben, « Thus we have, in a lucid presentation. 
93-1 > x 
gy = moni (15) 
23-1 ++ 
A, = À,_4 + a,ly, - 84,7) (16) 
À numerical simplification is possible by avoidine the divi- 
sion in eq. (15). It is obvious that {q, approaches a stafiona- 
ry value qq, which can be computed beforehand. Then , the 
convergent series (15) can be approximated by ambher series 
which converges against the same value Qoo but is easier to 
compute. In the present case we took recourse to the series 
133