Abstract:Inventory control typically considers controlling the price and the production rate. However, such systems have rigidity towards altering the physical storage capacity -- one can not easily alter the physical size after the initial design. The paper focuses on this critical aspect, consideration of which leads to a non-standard control problem. Here, the objective is a weighted combination of the classical integral term (formed by usual inventory costs) and an $L^{\infty}$ term (the maximum inventory level in the entire planning horizon). Our approach is to consider an additional state component to capture the `instantaneous' $L^{\infty}$ term (maximum inventory level till that instant) by virtue of which, we could convert the problem to the classical framework. For the direct ($L^{\infty}$) problem, we first identify a relation between the optimal price and the production rate policy, thereby reducing the dimensionality of the problem. By numerically solving a smooth variant of the converted problem, we obtain an optimal policy that illustrates a significant reduction in the storage capacity requirement. Interestingly, the loss in the revenue is negligible (less than $6\%$). As the importance of the $L^{\infty}$ component increases, the variations in the corresponding optimal inventory-level trajectory reduce. In the scenarios with partial/zero information about future demand curves, the above observation provides a guidance -- one should continually tune the policies to maintain instantaneous inventory-levels as close to zero as possible. With such a policy, the reduction in revenue is negligible, while having significant improvements for storage capacity. We theoretically establish certain interesting properties of the optimal policy, which also support the above guidance.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is a crucial aspect in inventory management, namely the design of physical storage capacity. Specifically, the paper focuses on how to optimally determine the maximum inventory storage capacity within a given planning period. This design problem is a non - standard control problem because traditional inventory control systems usually only consider the control of price and productivity rate, while ignoring the adjustment of physical storage capacity. Once the initial design is completed, the physical storage capacity is very difficult to change.
### Main contributions of the paper
1. **Introduction of new state variables**: In order to transform the problem into a classical framework, the paper introduces an auxiliary state variable \(y(t)\) to capture the "instantaneous" \(L^\infty\) term (i.e., the maximum inventory level up to time \(t\)). In this way, the \(L^\infty\) term can be converted into a terminal cost.
2. **Theoretical results**: The paper establishes some interesting properties of the optimal policy, for example:
- There is a linear relationship between the productivity rate policy and the price policy, which reduces the dimension of the problem.
- When there is an inventory surplus, the optimal productivity rate locally increases; when there is an inventory shortage, the optimal productivity rate locally decreases.
- The optimal inventory will not end with a surplus.
3. **Numerical analysis**: By numerically solving the smoothed approximation problem, the paper shows that as the importance of the \(L^\infty\) term increases, the change in the optimal inventory trajectory decreases, but the loss of revenue is very small (not more than 6%). For cases where future demand information is partially or completely unknown, it is recommended that manufacturers keep the immediate inventory level close to zero and allow a small shortage at the end of the planning period.
### Mathematical model
The mathematical model used in the paper includes the following main equations:
1. **Demand function**:
\[
d(t)=(\alpha(t)-\beta(t)p(t))^+
\]
where \(d(t)\) is the demand realized by the manufacturer at time \(t\), \(p(t)\) is the price quoted by the manufacturer at time \(t\), and \(\alpha(t)\) and \(\beta(t)\) represent the maximum possible demand and the price - sensitivity factor respectively.
2. **Inventory dynamics**:
\[
\dot{x}(t)=u(t)-(\alpha(t)-\beta(t)p(t))^+
\]
where \(x(t)\) is the inventory level at time \(t\) and \(u(t)\) is the productivity rate.
3. **Objective function**:
\[
J(0, x; u, p)=\int_{0}^{T}\left(p(s)(\alpha(s)-\beta(s)p(s))^+-au^2(s)-h(x(s))\right)ds + h_T(x(T))
\]
where \(h(x)\) and \(h_T(x)\) represent the inventory holding and shortage cost functions in the intermediate and terminal periods respectively.
4. **Optimization problem including the \(L^\infty\) term**:
\[
v_\sigma(0, x)=\sup_{(u, p)\in S}\left(J(0, x; u, p)-\sigma\sup_{s\in[0, T]}\{x(s)\}\right)
\]
where \(\sigma\) is the weight of the \(L^\infty\) term.
### Conclusion
By introducing new state variables and numerical methods, the paper successfully solves the non - standard control problem including the \(L^\infty\) term, providing new ideas and methods for the design of inventory management systems. In particular, the strategy proposed in the paper has important guiding significance in practical applications, especially when future demand information is unclear.