高级优化理论与方法(十)
Simplex
Given a basis B B B, x = [ x B T , 0 ] T x=[x_B^T,0]^T x=[xBT,0]T, x B = B T b x_B=B^Tb xB=BTb
Goal: new basis B ′ = B → a p , a q B'=B\rightarrow a_p,a_q B′=B→ap,aq
[ A , b ] = [ B , D , b ] ⇒ [ I , Y , y 0 ] [A,b]=[B,D,b]\Rightarrow[I,Y,y_0] [A,b]=[B,D,b]⇒[I,Y,y0]
[ A , b ] = [ B ′ , D ′ , b ] ⇒ [ I , Y ′ , y 0 ′ ] [A,b]=[B',D',b]\Rightarrow[I,Y',y_0'] [A,b]=[B′,D′,b]⇒[I,Y′,y0′]
y 0 = x B , y 0 ′ = x B ′ y_0=x_B,y_0'=x_B' y0=xB,y0′=xB′
注: [ B ′ , D ′ ] [B',D'] [B′,D′]是把原矩阵中的第 p p p列和第 q q q列交换后的结果。
Claim: y 1 j , ⋯ , y m j , m + 1 ≤ j ≤ n y_{1j},\cdots,y_{mj},m+1\leq j\leq n y1j,⋯,ymj,m+1≤j≤n, satisfy a j = y 1 j a 1 + ⋯ + y m j a m a_j=y_{1j}a_1+\cdots+y_{mj}a_m aj=y1ja1+⋯+ymjam
a q = y 1 q a 1 + ⋯ + y p q a p + ⋯ + y m q a m a_q=y_{1q}a_1+\cdots+y_{pq}a_p+\cdots+y_{mq}a_m aq=y1qa1+⋯+ypqap+⋯+ymqam
a p = a q y p q − ∑ i = 1 , i ≠ p m y i q y p q a i a_p=\frac{a_q}{y_{pq}}-\sum_{i=1,i\neq p}^m \frac{y_{iq}}{y_{pq}}a_i ap=ypqaq−∑i=1,i=pmypqyiqai
Claim: a 1 , ⋯ , a p − 1 , a p + 1 , a m , a q a_1,\cdots,a_{p-1},a_{p+1},a_m,a_q a1,⋯,ap−1,ap+1,am,aq linearly independent ⇔ y p q = 0 \Leftrightarrow y_{pq}=0 ⇔ypq=0
a p = y 1 q ′ a 1 + ⋯ + y p q ′ a q + ⋯ + y m q ′ a m a_p=y_{1q}'a_1+\cdots+y_{pq}'a_q+\cdots+y_{mq}'a_m ap=y1q′a1+⋯+ypq′aq+⋯+ymq′am
y p q ′ = 1 y p q , y i q ′ = − y i q y p q , ∀ 1 ≤ i ≤ m , i ≠ p y_{pq}'=\frac{1}{y_{pq}}, y_{iq}'=-\frac{y_{iq}}{y_{pq}},\forall 1\leq i \leq m, i\neq p ypq′=ypq1,yiq′=−ypqyiq,∀1≤i≤m,i=p
∀ j ≥ m + 1 , j ≠ q : { y i j − y p j y i q y p q , i ≠ p y p j y p q , i = p \forall j\geq m+1,j\neq q:\begin{cases} y_{ij}-\frac{y_{pj}y_{iq}}{y_{pq}},i\neq p \\ \frac{y_{pj}}{y_{pq}},i=p \end{cases} ∀j≥m+1,j=q:{yij−ypqypjyiq,i=pypqypj,i=p
y i 0 ′ = { y i 0 − y p 0 y i q y p q , i ≠ q y p 0 y p q , i = p y_{i0}'=\begin{cases} y_{i0}-\frac{y_{p0}y_{iq}}{y_{pq}},i\neq q\\ \frac{y_{p0}}{y_{pq}},i=p \end{cases} yi0′={yi0−ypqyp0yiq,i=qypqyp0,i=p
Claim: x ′ = [ y 0 ′ T , 0 ] T x'=[{y_0'}^T,0]^T x′=[y0′T,0]T satisfies A x = b Ax=b Ax=b
How to determine p and q?
determine p
Known q : ⇒ ? p q:\Rightarrow ?p q:⇒?p
basis: B , [ I , Y , y 0 ] B,[I,Y,y_0] B,[I,Y,y0]
a q = y 1 q a 1 + ⋯ + y m q a m a_q=y_{1q}a_1+\cdots+y_{mq}a_m aq=y1qa1+⋯+ymqam
{ ε a q = ε y 1 q a 1 + ⋯ + ε y m q a m b = y 10 a 1 + ⋯ + y m 0 a m \begin{cases} \varepsilon a_q=\varepsilon y_{1q}a_1+\cdots+\varepsilon y_{mq}a_m \\ b=y_{10}a_1+\cdots+y_{m0}a_m \end{cases} {εaq=εy1qa1+⋯+εymqamb=y10a1+⋯+ym0am
⇒ b = ( y 10 − ε y 1 q ) a 1 + ⋯ + ( y m 0 − ε y m q ) a m + ε a q \Rightarrow b=(y_{10}-\varepsilon y_{1q})a_1+\cdots+(y_{m0}-\varepsilon y_{mq})a_m+\varepsilon a_q ⇒b=(y10−εy1q)a1+⋯+(ym0−εymq)am+εaq
x ′ = [ y 10 − ε y 1 q , ⋯ , y m 0 − ε y m q , 0 , ⋯ , 0 , ε , 0 , ⋯ , 0 ] x'=[y_{10}-\varepsilon y_{1q},\cdots,y_{m0}-\varepsilon y_{mq},0,\cdots,0,\varepsilon,0,\cdots,0] x′=[y10−εy1q,⋯,ym0−εymq,0,⋯,0,ε,0,⋯,0]
A x ′ = b , x ′ ≥ 0 ⇒ ε = m i n { y i 0 y i q : y i q > 0 } ⇒ p = a r g m i n 1 ≤ i ≤ m { y i 0 y i q : y i q > 0 } Ax'=b,x'\geq 0\Rightarrow \varepsilon=min\{\frac{y_{i0}}{y_{iq}}:y_{iq}>0\}\Rightarrow p=argmin_{1\leq i\leq m}\{\frac{y_{i0}}{y_{iq}}:y_{iq}>0\} Ax′=b,x′≥0⇒ε=min{yiqyi0:yiq>0}⇒p=argmin1≤i≤m{yiqyi0:yiq>0}
new basis B ′ ⟷ a q a p B , p B' \stackrel{a_p}{\underset{a_q}\longleftrightarrow}B,p B′aq⟷apB,p satisfies y p 0 − ε y p q = 0 y_{p0}-\varepsilon y_{pq}=0 yp0−εypq=0
What if
y
i
q
≤
0
y_{iq}\leq 0
yiq≤0 for all
1
≤
i
≤
m
1\leq i\leq m
1≤i≤m?
⇒
\Rightarrow
⇒ for arbitrarily large
ε
≥
0
,
x
′
\varepsilon\geq 0,x'
ε≥0,x′ satisfies
A
x
′
=
b
,
x
′
≥
0
Ax'=b,x'\geq 0
Ax′=b,x′≥0
⇒
\Rightarrow
⇒ no solution [unbounded]
determine q
Goal: c T x ≥ c T x ′ c^Tx\geq c^Tx' cTx≥cTx′
Given B , x = [ y 10 , ⋯ , y m 0 , 0 , ⋯ , 0 ] T B,x=[y_{10},\cdots,y_{m0},0,\cdots,0]^T B,x=[y10,⋯,ym0,0,⋯,0]T
z 0 = c T x = c 1 y 10 + ⋯ + c m y m 0 z_0=c^Tx=c_1y_{10}+\cdots+c_my_{m0} z0=cTx=c1y10+⋯+cmym0
z = c T x ′ = c 1 ( y 10 − ε y 1 q ) + ⋯ + c m ( y m 0 − ε y m q ) + c q ε = z 0 − c p y p 0 + c q ε − ( c 1 ε y 1 q + ⋯ + c m ε y m q ) = z 0 − c p ε y p q + c q ε − ( c 1 ε y 1 q + ⋯ + c m ε y m q ) = z 0 + ε ( c q − ( c 1 y 1 q + ⋯ + c p y p q + ⋯ + c m y m q ) ) = z 0 + ε ( c q − z q ) z=c^Tx'\\ =c_1(y_{10}-\varepsilon y_{1q})+\cdots+c_m(y_{m0}-\varepsilon y_{mq})+c_q\varepsilon\\ =z_0-c_py_{p0}+c_q\varepsilon-(c_1\varepsilon y_{1q}+\cdots+c_m\varepsilon y_{mq})\\ =z_0-c_p\varepsilon y_{pq}+c_q\varepsilon-(c_1\varepsilon y_{1q}+\cdots+c_m\varepsilon y_{mq})\\ =z_0+\varepsilon(c_q-(c_1y_{1q}+\cdots+c_py_{pq}+\cdots+c_my_{mq}))\\ =z_0+\varepsilon(c_q-z_q) z=cTx′=c1(y10−εy1q)+⋯+cm(ym0−εymq)+cqε=z0−cpyp0+cqε−(c1εy1q+⋯+cmεymq)=z0−cpεypq+cqε−(c1εy1q+⋯+cmεymq)=z0+ε(cq−(c1y1q+⋯+cpypq+⋯+cmymq))=z0+ε(cq−zq)
⇒ z − z 0 = ε ( c q − z q ) = ε r q \Rightarrow z-z_0=\varepsilon(c_q-z_q)=\varepsilon r_q ⇒z−z0=ε(cq−zq)=εrq
∵
∀
m
+
1
≤
j
≤
n
:
c
j
−
z
j
\because \forall m+1\leq j \leq n: c_j-z_j
∵∀m+1≤j≤n:cj−zj is computable
∴
z
−
z
0
<
0
⇔
∃
m
+
1
≤
q
≤
n
:
r
q
=
c
q
−
z
q
<
0
\therefore z-z_0<0\Leftrightarrow \exist m+1\leq q\leq n: r_q=c_q-z_q<0
∴z−z0<0⇔∃m+1≤q≤n:rq=cq−zq<0
Claim: If ∀ m + 1 ≤ j ≤ n : r j = c j − z j ≥ 0 \forall m+1\leq j \leq n: r_j=c_j-z_j\geq 0 ∀m+1≤j≤n:rj=cj−zj≥0, then x x x optimal.
Algorithm
Simplex: Given an initial basis B
- Compute: [ A , b ] = [ B , D , b ] → [ I , Y , y 0 ] [A,b]=[B,D,b]\rightarrow [I,Y,y_0] [A,b]=[B,D,b]→[I,Y,y0]
- Compute: ∀ m + 1 ≤ j ≤ n : r j = c j − z j \forall m+1\leq j\leq n: r_j=c_j-z_j ∀m+1≤j≤n:rj=cj−zj
- If ∀ j : r j ≥ 0 \forall j: r_j\geq 0 ∀j:rj≥0, then return x = [ y 0 T , 0 ] T x=[y_0^T,0]^T x=[y0T,0]T
- else pick an q q q with r q < 0 r_q<0 rq<0
- If no
y
i
q
>
0
y_{iq}>0
yiq>0 for
1
≤
i
≤
m
1\leq i\leq m
1≤i≤m, then return “no solution”
else p = a r g m i n { y i 0 y i q : y i q > 0 } p=argmin\{\frac{y_{i0}}{y_{iq}}:y_{iq}>0\} p=argmin{yiqyi0:yiq>0} - Update [ I , Y , y 0 ] [I,Y,y_0] [I,Y,y0], goto 2
Example
min
−
x
1
−
2
x
2
−
x
3
-x_1-2x_2-x_3
−x1−2x2−x3
s.t.
2
x
1
+
x
2
−
x
3
≤
2
2
x
1
−
x
2
+
5
x
3
≤
6
4
x
1
+
x
2
+
x
3
≤
6
x
1
,
x
2
,
x
3
≥
0
2x_1+x_2-x_3\leq 2\\ 2x_1-x_2+5x_3\leq 6\\ 4x_1+x_2+x_3\leq 6\\ x_1,x_2,x_3\geq 0
2x1+x2−x3≤22x1−x2+5x3≤64x1+x2+x3≤6x1,x2,x3≥0
⇒ 2 x 1 + x 2 − x 3 + x 4 = 2 2 x 1 − x 2 + 5 x 3 + x 5 = 6 4 x 1 + x 2 + x 3 + x 6 = 6 x 1 , ⋯ , x 6 ≥ 0 \Rightarrow 2x_1+x_2-x_3+x_4=2\\ 2x_1-x_2+5x_3+x_5=6\\ 4x_1+x_2+x_3+x_6=6\\ x_1,\cdots,x_6\geq 0 ⇒2x1+x2−x3+x4=22x1−x2+5x3+x5=64x1+x2+x3+x6=6x1,⋯,x6≥0
A = [ 2 1 − 1 1 0 0 2 − 1 5 0 1 0 4 1 1 0 0 1 ] A=\begin{bmatrix} 2 &1&-1&1&0&0\\ 2&-1&5&0&1&0\\ 4&1&1&0&0&1 \end{bmatrix} A= 2241−11−151100010001
b = [ 2 6 6 ] b=\begin{bmatrix} 2\\ 6\\ 6 \end{bmatrix} b= 266
c = [ − 1 − 2 − 1 0 0 0 ] c=\begin{bmatrix} -1\\ -2\\ -1\\ 0\\ 0\\ 0 \end{bmatrix} c= −1−2−1000
B = [ 1 0 0 0 1 0 0 0 1 ] B=\begin{bmatrix} 1&0&0\\ 0&1&0\\ 0&0&1 \end{bmatrix} B= 100010001
[ x 4 x 5 x 6 x 1 x 2 x 3 b 1 0 0 2 1 − 1 2 0 1 0 2 − 1 5 6 0 0 1 4 1 1 6 ] \begin{bmatrix} x_4&x_5&x_6&x_1&x_2&x_3&b\\ 1&0&0&2&1&-1&2\\ 0&1&0&2&-1&5&6\\ 0&0&1&4&1&1&6 \end{bmatrix} x4100x5010x6001x1224x21−11x3−151b266
[ x 4 x 5 x 6 x 1 x 2 x 3 2 6 6 0 0 0 ] \begin{bmatrix} x_4&x_5&x_6&x_1&x_2&x_3\\ 2&6&6&0&0&0 \end{bmatrix} [x42x56x66x10x20x30]
z = [ 0 , 0 , 0 , − 1 , − 2 , − 1 ] [ 2 6 6 0 0 0 ] = 0 z=[0,0,0,-1,-2,-1]\begin{bmatrix} 2\\ 6\\ 6\\ 0\\ 0\\ 0 \end{bmatrix}=0 z=[0,0,0,−1,−2,−1] 266000 =0
z j = ∑ i = 1 m y i j c i = 0 z_j=\sum_{i=1}^m y_{ij}c_i=0 zj=∑i=1myijci=0 ,for 4 ≤ j ≤ 6 4\leq j \leq 6 4≤j≤6
r
j
=
c
j
−
z
j
=
{
−
1
,
j
=
4
−
2
,
j
=
5
−
1
,
j
=
6
r_j=c_j-z_j=\begin{cases} -1, j=4\\ -2,j=5\\ -1,j=6 \end{cases}
rj=cj−zj=⎩
⎨
⎧−1,j=4−2,j=5−1,j=6
⇒
q
=
5
\Rightarrow q=5
⇒q=5(取最小的)
y
i
0
y
i
q
=
{
2
,
i
=
1
/
,
i
=
2
6
,
i
=
3
\frac{y_{i0}}{y_{iq}}=\begin{cases} 2,i=1\\ /,i=2\\ 6,i=3 \end{cases}
yiqyi0=⎩
⎨
⎧2,i=1/,i=26,i=3
⇒
p
=
1
\Rightarrow p=1
⇒p=1(取正数中最小的)
y i j ′ = { y i j ′ − y p j y i q y p q , m + 1 ≤ j ≤ n , j ≠ q , i ≠ q y p j y p q , i = p y_{ij}'=\begin{cases} y_{ij}'-\frac{y_{pj}y_{iq}}{y_{pq}}, m+1\leq j\leq n,j\neq q,i\neq q\\ \frac{y_{pj}}{y_{pq}},i=p \end{cases} yij′={yij′−ypqypjyiq,m+1≤j≤n,j=q,i=qypqypj,i=p
y i q ′ = { 1 y p q , i = p − y i q y p q , i ≠ p y_{iq}'=\begin{cases} \frac{1}{y_{pq}},i=p\\ -\frac{y_{iq}}{y_{pq}},i\neq p \end{cases} yiq′={ypq1,i=p−ypqyiq,i=p
[ 1 0 0 2 1 − 1 2 0 1 0 4 1 3 8 0 0 1 2 − 1 2 4 ] \begin{bmatrix} 1&0&0&2&1&-1&2\\ 0&1&0&4&1&3&8\\ 0&0&1&2&-1&2&4 \end{bmatrix} 10001000124211−1−132284
B ′ = [ a 2 , a 5 , a 6 ] , Y ′ = [ a 1 , a 4 , a 3 ] B'=[a_2,a_5,a_6],Y'=[a_1,a_4,a_3] B′=[a2,a5,a6],Y′=[a1,a4,a3]
x ′ = [ x 2 x 5 x 6 x 1 x 4 x 3 2 8 4 0 0 0 ] x'=\begin{bmatrix} x_2&x_5&x_6&x_1&x_4&x_3\\ 2&8&4&0&0&0 \end{bmatrix} x′=[x22x58x64x10x40x30]
z ′ = − 4 z'=-4 z′=−4
c = [ − 2 , 0 , 0 , − 1 , 0 , − 1 ] c=[-2,0,0,-1,0,-1] c=[−2,0,0,−1,0,−1]
z j = { − 4 , j = 4 − 2 , j = 5 2 , j = 6 z_j=\begin{cases} -4, j=4\\ -2,j=5\\ 2,j=6 \end{cases} zj=⎩ ⎨ ⎧−4,j=4−2,j=52,j=6
r
j
=
c
j
−
z
j
=
{
3
,
j
=
4
2
,
j
=
5
−
3
,
j
=
6
r_j=c_j-z_j=\begin{cases} 3, j=4\\ 2,j=5\\ -3,j=6 \end{cases}
rj=cj−zj=⎩
⎨
⎧3,j=42,j=5−3,j=6
⇒
q
=
6
\Rightarrow q=6
⇒q=6
由于步骤比较繁琐,样例不再继续计算。
Initial Basic Feasible Solution
min
c
T
x
c^Tx
cTx
s.t.
A
x
=
b
Ax=b
Ax=b
x
≥
0
x\geq 0
x≥0
⇒
\Rightarrow
⇒ Artificial LP
min
y
1
+
⋯
+
y
m
≥
0
y_1+\cdots+y_m\geq 0
y1+⋯+ym≥0
s.t.
[
A
,
I
]
[
x
y
]
=
b
[A, I]\begin{bmatrix} x\\ y \end{bmatrix}=b
[A,I][xy]=b
x
≥
0
,
y
≥
0
x\geq 0,y\geq 0
x≥0,y≥0
Theorem
Thm: The initial LP has a basic feasible solution iff the artificial LP has an optimal solution with the objective function equal to 0.
Two-Phases Simplex
min
c
T
x
c^Tx
cTx
s.t.
A
x
=
b
Ax=b
Ax=b
x
≥
0
x\geq 0
x≥0
Assume
b
≥
0
b\geq 0
b≥0
Phase I {\rm I} I
- Construct artifitial LP
- Apply simplex to artifitial LP with an initial basic feasible solution [ 0 , ⋯ , 0 , b ] [0,\cdots,0,b] [0,⋯,0,b]
- If the optimal solution of the artifitial LP returns a value ∑ y i ≠ 0 \sum y_i\neq 0 ∑yi=0, then “no solution”
- else Phase I I {\rm II} II
Phase I I {\rm II} II
Apply Simplex to the initial LP with the optimal solution of the artificial LP.
Example
min
2
x
1
+
3
x
2
2x_1+3x_2
2x1+3x2
s.t.
4
x
1
+
2
x
2
−
x
3
=
12
4x_1+2x_2-x_3=12
4x1+2x2−x3=12
x
1
+
x
2
−
x
4
=
6
x_1+x_2-x_4=6
x1+x2−x4=6
x
1
,
⋯
,
x
4
≥
0
x_1,\cdots,x_4\geq 0
x1,⋯,x4≥0
Artifitial LP:
min
y
1
+
y
2
y_1+y_2
y1+y2
s.t.
4
x
1
+
2
x
2
−
x
3
+
y
1
=
12
4x_1+2x_2-x_3+y_1=12
4x1+2x2−x3+y1=12
x
1
+
x
2
−
x
4
+
y
2
=
6
x_1+x_2-x_4+y_2=6
x1+x2−x4+y2=6
x
1
,
x
2
,
x
3
,
x
4
,
y
1
,
y
2
≥
0
x_1,x_2,x_3,x_4,y_1,y_2\geq 0
x1,x2,x3,x4,y1,y2≥0
A = [ 4 2 − 1 0 1 0 12 1 1 0 − 1 0 1 6 ] A=\begin{bmatrix} 4&2&-1&0&1&0&12\\ 1&1&0&-1&0&1&6 \end{bmatrix} A=[4121−100−11001126]
x = 0 , y = [ 12 , 6 ] T , z = 18 x=0,y=[12,6]^T,z=18 x=0,y=[12,6]T,z=18
p = 5 , q = 2 p=5,q=2 p=5,q=2
A = [ 7 2 0 − 1 1 2 1 − 1 2 9 1 4 1 0 − 1 4 0 1 4 3 2 ] A=\begin{bmatrix} \frac{7}{2}&0&-1&\frac{1}{2}&1&-\frac{1}{2}&9\\ \frac{1}{4}&1&0&-\frac{1}{4}&0&\frac{1}{4}&\frac{3}{2} \end{bmatrix} A=[274101−1021−4110−2141923]
y 1 = 9 , x 2 = 3 2 , x 1 = x 3 = x 4 = y 2 = 0 , z = 9 y_1=9,x_2=\frac{3}{2},x_1=x_3=x_4=y_2=0,z=9 y1=9,x2=23,x1=x3=x4=y2=0,z=9
r
j
=
{
−
7
2
,
j
=
1
1
,
j
=
3
−
1
2
,
j
=
4
3
2
,
j
=
6
r_j=\begin{cases} -\frac{7}{2},j=1\\ 1,j=3\\ -\frac{1}{2},j=4\\ \frac{3}{2},j=6 \end{cases}
rj=⎩
⎨
⎧−27,j=11,j=3−21,j=423,j=6
⇒
q
=
1
\Rightarrow q=1
⇒q=1
y
i
0
y
i
q
=
{
2
,
i
=
1
6
,
i
=
2
\frac{y_{i0}}{y_{iq}}=\begin{cases} 2,i=1\\ 6,i=2 \end{cases}
yiqyi0={2,i=16,i=2
⇒
p
=
1
\Rightarrow p=1
⇒p=1
注:此处
p
=
1
,
q
=
1
p=1,q=1
p=1,q=1表示用
x
1
x_1
x1的那一列和
y
1
y_1
y1的那一列进行交换。
A = [ 1 0 − 2 7 1 7 2 7 − 1 7 18 7 0 1 1 14 − 2 7 − 1 14 2 7 6 7 ] A=\begin{bmatrix} 1&0&-\frac{2}{7}&\frac{1}{7}&\frac{2}{7}&-\frac{1}{7}&\frac{18}{7}\\ 0&1&\frac{1}{14}&-\frac{2}{7}&-\frac{1}{14}&\frac{2}{7}&\frac{6}{7} \end{bmatrix} A=[1001−7214171−7272−141−717271876]
x 1 = 18 7 , x 2 = 6 7 , x 3 = x 4 = y 1 = y 2 = 0 , z = 0 x_1=\frac{18}{7},x_2=\frac{6}{7},x_3=x_4=y_1=y_2=0,z=0 x1=718,x2=76,x3=x4=y1=y2=0,z=0
注:此时, z = 0 z=0 z=0表示第一阶段已经达到了最优,接下来开始做第二阶段。
[ 1 0 − 2 7 1 7 18 7 0 1 1 14 − 2 7 6 7 ] \begin{bmatrix} 1&0&-\frac{2}{7}&\frac{1}{7}&\frac{18}{7}\\ 0&1&\frac{1}{14}&-\frac{2}{7}&\frac{6}{7} \end{bmatrix} [1001−7214171−7271876]
x = [ 18 7 , 6 7 , 0 , 0 ] , z = 8 x=[\frac{18}{7},\frac{6}{7},0,0],z=8 x=[718,76,0,0],z=8
总结
本节课着重解决上次课提出的关于线性规划的三个问题,即如何找到初始可行解、如何确定 x x x达到最优、如何从 x x x构造新的可行解。这节课通过一系列的推导,推导出了单纯形法,并给出了单纯形法的算法和样例。对于如何寻找初始可行解的问题,又提出了两阶段的单纯形法。下节课将讨论单纯形法的时间复杂度以及一些优化。