A decision before demand is known
A shop must order a nonnegative integer number of items before a single period's demand is observed. Ordering too few misses sales; ordering too many leaves surplus. A model makes the trade-off explicit rather than assuming that average demand is always the right order.
Definition 8.1.1
(The single-period order model)
is a nonnegative integer random variable with a known distribution function . Each item costs , each sold item earns , and every unsold item is salvaged for , where . For demand and order the profit is
where . There is no fixed ordering cost, shortage penalty, stock carried to another period, or capacity constraint. Demand exceeding stock is lost, not backordered. The goal is to maximize expected single-period profit , not to guarantee a profit for every demand realization. The number is the critical ratio.
One more unit
Theorem 8.1.2
(Expected value of one more unit)
For every nonnegative integer ,
Proof
(Splitting on whether the extra unit sells)
Compare order quantities and on the same demand outcome. If , the extra unit is sold and changes profit by . If , it is salvaged and changes profit by . These events partition the integer-valued demand outcomes. Taking expectations gives the first formula. Using and gives the second.
The increments are nonincreasing because is nondecreasing. Thus adding a unit helps while , ties at equality, and hurts after the inequality reverses.
Theorem 8.1.3
(The first crossing of the critical ratio is optimal)
Let be the smallest nonnegative integer with . Then exists and maximizes expected profit over all nonnegative integer orders.
Proof
(Positive increments before the crossing, nonpositive after)
Since , the critical ratio lies strictly between zero and one, and a probability distribution on nonnegative integers has , so exists. For , is below the ratio and ; for , and . For any order , the difference is a sum of increments between and , each positive (if ) or each nonpositive and subtracted (if ). In both cases the difference is nonnegative.
Proof X-ray
Monotone increments do the work of concavity
The proof never compares with a distant order directly. It uses the fact that the increments are nonincreasing, which follows from being a distribution function, and then telescopes. A sequence whose increments change sign once, from positive to nonpositive, is maximized where the sign changes; this is the discrete form of concavity.
Each cost assumption has a specific job. The condition greater than makes the ratio positive, so ordering is ever worthwhile. The condition greater than makes the ratio below one, which is what guarantees a crossing: if salvage equaled cost, the ratio would be one and for unbounded demand would never reach it, so every increment would be positive and no order would be optimal. Ties are allowed by the inequality, which is why other optimal orders can appear where an increment is exactly zero.
A concrete distribution
Worked example 8.1.4
(The first crossing for a three-point demand)
Let , , , and let equal with probabilities . The critical ratio is . Here and , so the first crossing is .
At profit is always zero. At it is when and otherwise, so expected profit is . The increment from to is , giving expected profit at . Further units are always salvaged at a loss of two each. The expected demand happens to be one in this example, but that coincidence is not the proof of optimality.
Counterexample mode
The mean demand need not be the best order
Keep , , , but let with probability and with probability . Mean demand is one, yet exceeds the ratio , so the first crossing is . Ordering one has expected increment , hence is worse than ordering nothing. The distribution's shape matters, not just its mean.
Worked example 8.1.5
(A tie at the critical ratio)
Keep , , and let be uniform on . Then , , , so and . Directly, gives and gives . Both are optimal; rules out .
Worked example 8.1.6
(A higher margin and a capacity limit)
Return to the three-point demand with , , . The ratio is . Since and , the first crossing is , with . Expected profits are at and at . If capacity is , the best feasible order is : the increment before the crossing is positive, so the capacity binds.
Common mistakes
Common mistake
(Leaving salvage out of the critical ratio)
The threshold is , not . With , , and the three-point demand, is already reached by and would suggest , while the correct ratio gives and expected profit instead of .
Common mistake
(An off-by-one in the selling probability)
The -st unit sells exactly when . Using instead at in the first example gives and suggests ordering a second unit, whereas the correct increment is .
Limits and extensions
The exact threshold changes if lost demand incurs a penalty, salvage depends on quantity, ordering has a fixed cost, or the decision is continuous. A nonnegative integer capacity bound also changes the feasible set: under the stated model the optimal feasible integer order is the first-crossing order if it fits, and otherwise the capacity, because increments before the crossing are positive. Estimating an unknown demand distribution is a separate statistical task; knowing a formula does not make its input reliable.
Exercises
Quick check
(In the three-point example with , calculate the expected profit for directly from the three demand outcomes.)
Solution · Answer
At , profits are for , respectively. Their expectation is .
Quick check
(Suppose and with , , . What can be said about and ?)
Solution · Answer
is the first crossing, hence optimal, and , so is also optimal. Further quantities tie only if more increments are zero.
Quick check
(Keep the three-point example with but impose capacity zero. What is the best feasible order and its expected profit?)
Solution · Answer
Only is feasible, giving expected profit .
Quick check
(Why does the mean-demand counterexample have mean one even though ordering one is suboptimal?)
Solution · Answer
The mean is , while the first unit is sold with probability only , so its expected increment is .