Abstract:We consider a class of optimization problems defined by a system of linear equations with min and max operators. This class of optimization problems has been studied under restrictive conditions, such as, (C1) the halting or stability condition; (C2) the non-negative coefficients condition; (C3) the sum up to 1 condition; and (C4) the only min or only max oerator condition. Several seminal results in the literature focus on special cases. For example, turn-based stochastic games correspond to conditions C2 and C3; and Markov decision process to conditions C2, C3, and C4. However, the systematic computational complexity study of all the cases has not been explored, which we address in this work. Some highlights of our results are: with conditions C2 and C4, and with conditions C3 and C4, the problem is NP-complete, whereas with condition C1 only, the problem is in UP intersects coUP. Finally, we establish the computational complexity of the decision problem of checking the respective conditions.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the computational complexity classification of linear equation systems containing minimum and maximum operators. Specifically, the authors study a class of optimization problems defined by linear equations with minimum (min) and maximum (max) operators. Such optimization problems have been studied under specific constraints, for example:
- **Condition C1**: Stopping or stability condition
- **Condition C2**: Non - negative coefficient condition
- **Condition C3**: The sum - is - 1 condition
- **Condition C4**: The condition of containing only minimum or only maximum operators
However, the systematic study of the computational complexity under all these combinations of conditions has not been fully explored. Therefore, this paper aims to fill this gap and provide a comprehensive computational complexity analysis.
### Main contributions
1. **NP - completeness results**:
- When conditions C2 and C4 are satisfied, or conditions C3 and C4 are satisfied, the problem is NP - complete.
2. **UP ∩ COUP results**:
- Only when condition C1 is satisfied, the problem belongs to UP ∩ COUP.
3. **SSG - hardness**:
- Even when conditions C1, C3 and C4 are satisfied, the problem is still SSG - hard, which means that proving the existence of a polynomial - time algorithm requires a major breakthrough.
### Summary of computational complexity
Table 1 summarizes the computational complexity of the LEMM decision problem under different combinations of conditions:
| Combinations of conditions | Complexity |
| --- | --- |
| {C2, C4}, {C3, C4}, {C2}, {C3}, {C4}, ∅ | NP - complete |
| {C1, C3, C4}, {C1, C3}, {C1, C4}, {C1} | UP ∩ COUP (SSG - hard) |
| {C1, C2}, {C1, C2, C3} | UP ∩ COUP |
| {C1, C2, C3, C4}, {C2, C3, C4}, {C1, C2, C4} | PTIME |
In addition, the authors also study the computational complexity of verifying these conditions themselves, as shown in Table 2.
### Technical contributions
The authors demonstrate their technical contributions in the following ways:
1. **Failure of classical properties**:
- When only condition C1 is satisfied and conditions C2 and C3 are not satisfied, some classical SSG properties (such as monotonicity, minimax = maxmin, etc.) no longer hold. The authors provide specific examples to illustrate this.
2. **Existence of a unique solution**:
- Even when the classical properties fail, the authors still prove the existence of a unique solution under condition C1, thus establishing the UP ∩ COUP result.
### Conclusion
This paper fills the gaps in the existing literature by systematically studying the computational complexity of linear equation systems with minimum and maximum operators, and provides new complexity results and technical insights.