the partial derivatives are computed as above and we get finally:
s
2
y
i
; y
j
¼
X
klmn
ik
y
l
jm
y
n
s
2
a
kl
a
mn
þ
X
kl
ik
jl
s
2
x
k
x
l
þ
X
klmn
ð
ik
y
l
jm
jk
y
l
im
Þs
2
a
kl
x
m
ð7:79Þ
which simplifies, if the variables are independent (that is, if the covariances are
zero, which is not always the case):
s
2
y
i
; y
j
¼
X
kl
ik
jk
y
2
l
n
s
2
a
ki
a
kl
þ
X
kl
ik
jk
s
2
x
k
x
k
ð7:80Þ
Upper bound of the errors
The vector y contains a large number of data, but it is helpful to represent the
error by a single value. To obtain such a single value, the following defini tions,
which can be found in the specific mathematical literature (for example, Deif,
1986) are used.
Vectorial norms and matrix norms
The norm jxj of a vector x is any operation of the n-fold real space R
n
in the
ensemble of real numbers R satisfying:
jxj0 and jxj¼0 if and only if x ¼ 0
jcxj¼jcjjxj for any c 2 R
jx þ yjjxjþjyj
ð7:81Þ
For example, the Euclidian norm that corresponds best to the standard
deviation:
jxj
2
¼
ffiffiffiffiffiffiffiffiffiffiffiffiffi
X
x
2
i
q
ð7:82Þ
complies with the relations in Equation 7.81, but there are many others, like
jxj
1
¼
P
x
i
or j xj
1
¼ maxðjx
i
jÞ.
The norm jAj of a matrix A is any application NðAÞ)jAj2R satisfying:
jAj0 and jAj¼0 if and only if A ¼ 0
jcAj¼jcjjAj for any c 2 R
jA þ BjjAjþjBj
jA BjjAj jBj
ð7:83Þ
The matrix norm jAj is consistent with the vectorial norm jxj if:
jAxjjAj jxj for any x ð7:84Þ
162 Ventilation and Airflow in Buildings
Kommentare zu diesen Handbüchern