考慮一組
(
x
1
,
y
1
)
,
(
x
2
,
y
2
)
,
…
,
(
x
m
,
y
m
)
{\displaystyle (x_{1},y_{1}),(x_{2},y_{2}),\dots ,(x_{m},y_{m})}
m
{\displaystyle m}
y
^
=
f
(
x
,
β
)
{\displaystyle {\hat {y}}=f(x,{\boldsymbol {\beta }})}
x
β
=
(
β
1
,
β
2
,
…
,
β
n
)
{\displaystyle {\boldsymbol {\beta }}=(\beta _{1},\beta _{2},\dots ,\beta _{n})}
n
m
≥
n
{\displaystyle m\geq n}
β
{\displaystyle {\boldsymbol {\beta }}}
S
=
∑
i
=
1
m
r
i
2
,
{\displaystyle S=\sum _{i=1}^{m}r_{i}^{2},}
其中殘差 ri
r
i
=
y
i
−
f
(
x
i
,
β
)
,
(
i
=
1
,
2
,
…
,
m
)
.
{\displaystyle r_{i}=y_{i}-f(x_{i},{\boldsymbol {\beta }}),\qquad (i=1,2,\dots ,m).}
S 最小值 時的梯度 為零。由於模型包含n n
∂
S
∂
β
j
=
2
∑
i
r
i
∂
r
i
∂
β
j
=
0
(
j
=
1
,
…
,
n
)
.
{\displaystyle {\frac {\partial S}{\partial \beta _{j}}}=2\sum _{i}r_{i}{\frac {\partial r_{i}}{\partial \beta _{j}}}=0\quad (j=1,\ldots ,n).}
在非線性系統中，偏導數
∂
r
i
∂
β
j
{\textstyle {\frac {\partial r_{i}}{\partial \beta _{j}}}}
x
β
{\displaystyle {\boldsymbol {\beta }}}
β
j
≈
β
j
k
+
1
=
β
j
k
+
Δ
β
j
.
{\displaystyle \beta _{j}\approx \beta _{j}^{k+1}=\beta _{j}^{k}+\Delta \beta _{j}.}
其中，k
Δ
β
{\displaystyle \Delta {\boldsymbol {\beta }}}
β
k
{\displaystyle {\boldsymbol {\beta }}^{k}}
泰勒級數 展開以線性化模型：
f
(
x
i
,
β
)
≈
f
(
x
i
,
β
k
)
+
∑
j
∂
f
(
x
i
,
β
k
)
∂
β
j
(
β
j
−
β
j
k
)
=
f
(
x
i
,
β
k
)
+
∑
j
J
i
j
Δ
β
j
.
{\displaystyle f(x_{i},{\boldsymbol {\beta }})\approx f(x_{i},{\boldsymbol {\beta }}^{k})+\sum _{j}{\frac {\partial f(x_{i},{\boldsymbol {\beta }}^{k})}{\partial \beta _{j}}}\left(\beta _{j}-\beta _{j}^{k}\right)=f(x_{i},{\boldsymbol {\beta }}^{k})+\sum _{j}J_{ij}\,\Delta \beta _{j}.}
雅可比矩陣 J J
∂
r
i
∂
β
j
=
−
J
i
j
,
{\displaystyle {\frac {\partial r_{i}}{\partial \beta _{j}}}=-J_{ij},}
殘差的表達式則為
Δ
y
i
=
y
i
−
f
(
x
i
,
β
k
)
,
{\displaystyle \Delta y_{i}=y_{i}-f(x_{i},{\boldsymbol {\beta }}^{k}),}
r
i
=
y
i
−
f
(
x
i
,
β
)
=
(
y
i
−
f
(
x
i
,
β
k
)
)
+
(
f
(
x
i
,
β
k
)
−
f
(
x
i
,
β
)
)
≈
Δ
y
i
−
∑
s
=
1
n
J
i
s
Δ
β
s
.
{\displaystyle r_{i}=y_{i}-f(x_{i},{\boldsymbol {\beta }})=\left(y_{i}-f(x_{i},{\boldsymbol {\beta }}^{k})\right)+\left(f(x_{i},{\boldsymbol {\beta }}^{k})-f(x_{i},{\boldsymbol {\beta }})\right)\approx \Delta y_{i}-\sum _{s=1}^{n}J_{is}\Delta \beta _{s}.}
將上述表達式代入梯度方程，可以得到
−
2
∑
i
=
1
m
J
i
j
(
Δ
y
i
−
∑
s
=
1
n
J
i
s
Δ
β
s
)
=
0
,
{\displaystyle -2\sum _{i=1}^{m}J_{ij}\left(\Delta y_{i}-\sum _{s=1}^{n}J_{is}\ \Delta \beta _{s}\right)=0,}
以上方程可化簡為n 正規方程 （normal equations）：
∑
i
=
1
m
∑
s
=
1
n
J
i
j
J
i
s
Δ
β
s
=
∑
i
=
1
m
J
i
j
Δ
y
i
(
j
=
1
,
…
,
n
)
.
{\displaystyle \sum _{i=1}^{m}\sum _{s=1}^{n}J_{ij}J_{is}\ \Delta \beta _{s}=\sum _{i=1}^{m}J_{ij}\ \Delta y_{i}\qquad (j=1,\dots ,n).}
正規方程可用矩陣表示法寫成
(
J
T
J
)
Δ
β
=
J
T
Δ
y
.
{\displaystyle \left(\mathbf {J} ^{\mathsf {T}}\mathbf {J} \right)\Delta {\boldsymbol {\beta }}=\mathbf {J} ^{\mathsf {T}}\ \Delta \mathbf {y} .}
上述方程是使用高斯-牛頓算法
需要注意的是雅可比矩陣定義中導數的符號約定。某些文獻中的J
不同數據點（觀測結果）的可靠性並不一定相同，此時可使用加權平方和
S
=
∑
i
=
1
m
W
i
i
r
i
2
.
{\displaystyle S=\sum _{i=1}^{m}W_{ii}r_{i}^{2}.}
權重矩陣W 對角矩陣 ，理想情況下每個權重係數應等於觀測誤差方差 的倒數。[ 1]
(
J
T
W
J
)
Δ
β
=
J
T
W
Δ
y
.
{\displaystyle \left(\mathbf {J} ^{\mathsf {T}}\mathbf {WJ} \right)\Delta {\boldsymbol {\beta }}=\mathbf {J} ^{\mathsf {T}}\mathbf {W} \ \Delta \mathbf {y} .}
