4.3 Consider an inventory problem similar to the problem of Section 4.2 (zero fixed cost). The only difference is that at the beginning of each period
k
k
k the decision maker, in addition to knowing the current inventory level
x
k
x_k
xk, receives an accurate forecast that the demand
w
k
w_k
wk will be selected in accordance with one out of two possible probability distributions
P
ℓ
,
P
s
P_{\ell},P_s
Pℓ,Ps (large demand, small demand). The a priori probability of a large demand forecast is known (d. Section 1.4).
(1) Obtain the optimal ordering policy for the case of a single-period problem.
(2) Extend the result to the
N
N
N-period case.
(3) Extend the result to the case of any finite number of possible distributions.
Solution. (1) Assume the priori probability of
P
ℓ
,
P
s
P_{\ell},P_s
Pℓ,Ps is
 
p
ℓ
,
p
s
\,p_{\ell},p_s
pℓ,ps respectively.
J
1
(
x
1
,
y
1
)
=
0
J_1(x_1,y_1)=0
J1(x1,y1)=0
J
0
(
x
0
,
y
0
)
=
p
s
⋅
min
u
0
[
c
u
0
+
H
(
x
0
+
u
0
−
w
0
)
]
+
p
ℓ
⋅
min
u
0
[
c
u
0
+
H
(
x
0
+
u
0
−
w
0
)
]
J_0(x_0,y_0)=p_s\cdot\min_{u_0}\left[cu_0+H(x_0+u_0-w_0)\right]+p_{\ell}\cdot\min_{u_0}\left[cu_0+H(x_0+u_0-w_0)\right]
J0(x0,y0)=ps⋅u0min[cu0+H(x0+u0−w0)]+pℓ⋅u0min[cu0+H(x0+u0−w0)]
=
c
u
0
+
p
⋅
p
s
(
x
0
+
u
0
−
w
0
)
−
h
⋅
p
ℓ
(
x
0
+
u
0
−
w
0
)
=cu_0+p\cdot p_s(x_0+u_0-w_0)-h\cdot p_{\ell}(x_0+u_0-w_0)\qquad\qquad
=cu0+p⋅ps(x0+u0−w0)−h⋅pℓ(x0+u0−w0)
=
c
u
0
+
⋅
(
p
⋅
p
s
−
h
⋅
p
ℓ
)
(
x
0
+
u
0
−
w
0
)
          
=cu_0+\cdot (p\cdot p_s-h\cdot p_{\ell})(x_0+u_0-w_0)\qquad\qquad\;\;\;\;\;\qquad\qquad
=cu0+⋅(p⋅ps−h⋅pℓ)(x0+u0−w0)
?
?
?
?
?
?
?
?
?
?
?
?
?
?????????????
?????????????
4.5 Consider the inventory problem of Section 4.2 for the case where the cost has the general form
E
{
∑
k
=
0
N
r
k
(
x
k
)
}
\mathbf{E}\left\{\sum_{k=0}^Nr_k(x_k)\right\}
E{k=0∑Nrk(xk)} The functions
r
k
r_k
rk are convex and differentiable and
lim
x
→
−
∞
d
r
(
x
)
d
x
=
−
∞
,
lim
x
→
∞
d
r
(
x
)
d
x
=
∞
k
=
0
,
⋯
 
,
N
\lim_{x\rightarrow -\infty}\frac{d r(x)}{dx}=-\infty,\qquad \lim_{x\rightarrow \infty}\frac{d r(x)}{dx}=\infty\qquad k=0,\cdots,N
x→−∞limdxdr(x)=−∞,x→∞limdxdr(x)=∞k=0,⋯,N (1) Assume that the fixed cost is zero. Write the DP algorithm for this problem and show that the optimal ordering policy has the same form as the one derived in Section 4.2.
(2) Suppose there is a one-period time lag between the order and the delivery of inventory; that is, the system equation is of the form
x
k
+
1
=
x
k
+
u
k
−
1
−
w
k
,
k
=
0
,
1
,
⋯
 
,
N
−
1
x_{k+1}=x_k+u_{k-1}-w_k,\qquad k=0,1,\cdots,N-1
xk+1=xk+uk−1−wk,k=0,1,⋯,N−1 where
u
−
1
u_{-1}
u−1 is given. Reformulate the problem so that it has the form of the problem of part (1).
Solution.
4.13 (A Gambling Problem) A gambler enters a game whereby he may at time k k k stake any amount u l ≥ 0 u_l\ge 0 ul≥0 that does not exceed his current fortune x k x_k xk (defined to be his initial capital plus his gain or minus his loss thus far). He wins his stake back and as much more with probability p p p, 0 < p < 1 0<p<1 0<p<1, and he loses his stake with probability ( 1 − p ) (1-p) (1−p). Show that the gambling strategy that maximizes E { ln x N } \mathbf{E}\{\ln x_N\} E{lnxN}, where x N x_N xN denotes his fortune after N N N plays, is to stake at each time k k k an amount u k = ( 2 p − 1 ) x k u_k=(2p-1)x_k uk=(2p−1)xk.
Solution. The gambling system can be formulated as follows: x k + 1 = x k + w k u k x_{k+1}=x_k+w_ku_k xk+1=xk+wkuk where w k w_k wk is a random variable with probability p p p taking value 1 1 1, and probability ( 1 − p ) (1-p) (1−p) taking value − 1 -1 −1. And the DP-algorithm can be written as: J N ( x N ) = ln ( x N ) J_N(x_N)=\ln (x_N) JN(xN)=ln(xN) J k ( x k ) = min u k E w k [ J k + 1 ( x k + w k u k ) ] J_k(x_k)=\mathop{\min}\limits_{u_k}\mathop\mathbf{E}\limits_{w_k}\left[J_{k+1}(x_k+w_ku_k)\right] Jk(xk)=ukminwkE[Jk+1(xk+wkuk)] For k = N − 1 k=N-1 k=N−1, we have J N − 1 ( x N − 1 ) = min u k E w k { ln [ x N − 1 + w N − 1 x N − 1 ] } J_{N-1}(x_{N-1})=\mathop{\min}\limits_{u_k}\mathop\mathbf{E}\limits_{w_k}\left\{\ln[x_{N-1}+w_{N-1}x_{N-1}]\right\}\qquad\qquad\qquad\qquad\qquad\qquad JN−1(xN−1)=ukminwkE{ln[xN−1+wN−1xN−1]} = p ⋅ ln [ x N − 1 + u N − 1 ] + ( 1 − p ) ln ( x N − 1 − u N − 1 ) =p\cdot \ln[x_{N-1}+u_{N-1}]+(1-p)\ln(x_{N-1}-u_{N-1}) =p⋅ln[xN−1+uN−1]+(1−p)ln(xN−1−uN−1) Calculate the derivative at u N − 1 u_{N-1} uN−1 of the right hand side, we obtain p x N − 1 + u N − 1 − 1 − p x N − 1 − u N − 1 = ( 2 p − 1 ) x N − 1 − u N − 1 x N − 1 2 − u N − 1 2 = 0 \frac{p}{x_{N-1}+u_{N-1}}-\frac{1-p}{x_{N-1}-u_{N-1}}=\frac{(2p-1)x_{N-1}-u_{N-1}}{x_{N-1}^2-u_{N-1}^2}=0 xN−1+uN−1p−xN−1−uN−11−p=xN−12−uN−12(2p−1)xN−1−uN−1=0 yielding u N − 1 ∗ = ( 2 p − 1 ) x N − 1 u_{N-1}^*=(2p-1)x_{N-1} uN−1∗=(2p−1)xN−1and J N − 1 ∗ = p ln ( 2 p ) + ( 1 − p ) ln ( 2 − 2 p ) + ln x N − 1 = C p + ln ( 2 − 2 p ) J_{N-1}^*=p\ln(2p)+(1-p)\ln(2-2p)+\ln x_{N-1}=C_p+\ln(2-2p) JN−1∗=pln(2p)+(1−p)ln(2−2p)+lnxN−1=Cp+ln(2−2p) Using the similar augement and by induction, we can prove that for any k = 0 , 1 , ⋯   , N − 1 k=0,1,\cdots,N-1 k=0,1,⋯,N−1, u k ∗ = ( 2 p − 1 ) x k u_k^*=(2p-1)x_k uk∗=(2p−1)xk and J k ∗ ( x k ) = C k + ln x k J_k^*(x_k)=C_k+\ln x_k Jk∗(xk)=Ck+lnxk for some constant C k C_k Ck. □ \qquad\qquad\qquad\qquad\qquad\qquad\Box □