Computational complexity of the recoverable robust shortest path problem in acyclic digraphs

Adam Kasperski,Pawel Zielinski

2024-10-12

Abstract:In this paper, the recoverable robust shortest path problem in acyclic digraphs is considered. The interval budgeted uncertainty representation is used to model the uncertain second-stage costs. The computational complexity of this problem has been open to date. In this paper, we prove that the problem is strongly NP-hard even for the case of layered acyclic digraphs. We also show that for the discrete budgeted uncertainty, the problem is not approximable unless P=NP.

Data Structures and Algorithms,Optimization and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is **the computational complexity of the Recoverable Robust Shortest Path Problem (Rec Rob SP) in acyclic digraphs**. Specifically, this problem takes into account the second - stage costs of uncertainty and uses the interval - budget uncertainty representation to model these uncertainties.

### Problem Background and Definition

1. **Problem Description**:
- Given a directed acyclic graph \(G=(V, A)\), where \(|V| = n\) and \(|A| = m\).
- Designate two nodes \(s\in V\) and \(t\in V\) as the source and sink nodes respectively.
- For each arc \(e\in A\), specify the first - stage cost \(C_e\geq0\).
- The second - stage costs are uncertain, and it is assumed that they belong to an uncertainty set \(U\subseteq\mathbb{R}^m_+\).
- In the Recoverable Robust Shortest Path Problem, first calculate a first - stage path \(X\in\Phi\), and then after the second - stage cost scenario \(ccc\in U\) is revealed, the path \(X\) can be modified to a certain extent, and a new path \(Y\) is selected within a certain predetermined neighborhood \(\Phi(X, k)\).

2. **Neighborhood Definition**:
- Arc - inclusion neighborhood: \(\Phi_{\text{incl}}(X, k)=\{Y\in\Phi:|Y\setminus X|\leq k\}\)
- Arc - exclusion neighborhood: \(\Phi_{\text{excl}}(X, k)=\{Y\in\Phi:|X\setminus Y|\leq k\}\)
- Arc - symmetric - difference neighborhood: \(\Phi_{\text{sym}}(X, k)=\{Y\in\Phi:|(Y\setminus X)\cup(X\setminus Y)|\leq k\}\)

3. **Objective Function**:
- Minimize the total cost: \(\min_{X\in\Phi}\left(\sum_{e\in X}C_e+\max_{ccc\in U}\min_{Y\in\Phi(X, k)}\sum_{e\in Y}c_e\right)\)

### Main Contributions

- **Computational Complexity**:
- It is proved that even for layered directed acyclic graphs, this problem is strongly NP - hard.
- For discrete - budget uncertainty, it is proved that this problem cannot be approximated and solved under the condition of \(P\neq NP\).

### Proof Method

The author proves its complexity by reducing the Hamiltonian Path problem to the Rec Rob SP problem. The specific steps are as follows:

1. **Construct Instance**:
- Given an instance \(G = (V, A)\) of the Hamiltonian Path problem, construct a corresponding instance \(G'\) of the Rec Rob SP problem.
- \(G'\) contains multiple copy nodes and additional source node \(s\) and sink node \(t\), and these nodes are connected by vertical and diagonal arcs.

2. **Parameter Setting**:
- Set the budget parameters \(\Gamma_c = 1\) or \(\Gamma_d=n - 1\), and the recovery parameter \(k\).
- Set the first - stage cost of arcs to 0, the second - stage cost of vertical arcs within the interval \([0, M]\), and the second - stage cost of diagonal arcs within the interval \([M, M]\)