Revisiting Tardos's Framework for Linear Programming: Faster Exact Solutions using Approximate Solvers

Daniel Dadush,Bento Natura,László A. Végh

DOI: https://doi.org/10.48550/arXiv.2009.04942

2020-09-10

Abstract:In breakthrough work, Tardos (Oper. Res. '86) gave a proximity based framework for solving linear programming (LP) in time depending only on the constraint matrix in the bit complexity model. In Tardos's framework, one reduces solving the LP $\min \langle c,{x}\rangle$, $Ax=b$, $x \geq 0$, $A \in \mathbb{Z}^{m \times n}$, to solving $O(nm)$ LPs in $A$ having small integer coefficient objectives and right-hand sides using any exact LP algorithm. This gives rise to an LP algorithm in time poly$(n,m\log\Delta_A)$, where $\Delta_A$ is the largest subdeterminant of $A$. A significant extension to the real model of computation was given by Vavasis and Ye (Math. Prog. '96), giving a specialized interior point method that runs in time poly$(n,m,\log\bar\chi_A)$, depending on Stewart's $\bar{\chi}_A$, a well-studied condition number. In this work, we extend Tardos's original framework to obtain such a running time dependence. In particular, we replace the exact LP solves with approximate ones, enabling us to directly leverage the tremendous recent algorithmic progress for approximate linear programming. More precisely, we show that the fundamental "accuracy" needed to exactly solve any LP in $A$ is inverse polynomial in $n$ and $\log\bar{\chi}_A$. Plugging in the recent algorithm of van den Brand (SODA '20), our method computes an optimal primal and dual solution using ${O}(m n^{\omega+1} \log (n)\log(\bar{\chi}_A+n))$ arithmetic operations, outperforming the specialized interior point method of Vavasis and Ye and its recent improvement by Dadush et al (STOC '20). At a technical level, our framework combines together approximate LP solutions to compute exact ones, making use of constructive proximity theorems -- which bound the distance between solutions of "nearby" LPs -- to keep the required accuracy low.

Optimization and Control,Data Structures and Algorithms

What problem does this paper attempt to address?

This paper aims to solve several key problems in linear programming (LP), especially how to use approximate solvers to achieve exact solutions depending on the condition number of the matrix. Specifically, the goals of the paper are as follows:

1. **Improve the Tardos framework**: The paper proposes an improved Tardos framework that can obtain exact solutions by using an approximate linear programming solver depending on the matrix condition number \(\bar{\chi}_A\). The main difference from the original Tardos framework is that the new framework allows for dynamic adjustment of the perturbations of the objective function and the right - hand - side vector, rather than just pre - fixing these perturbations.

2. **Reduce computational complexity**: By combining the latest advances in approximate solvers, the method proposed in the paper is superior to existing methods in terms of computational complexity. In particular, the algorithm in the paper achieves a time complexity of \(O(mn^{\omega + 1}\log^2(n)\log(\bar{\chi}_A + n))\), where \(\omega\) is the exponent of matrix multiplication.

3. **Handle infeasibility**: The paper not only focuses on solving optimization problems but also provides methods for handling infeasibility problems. For infeasible linear programming problems, the paper provides an effective method to generate Farkas certificates to prove the infeasibility of the problem.

### Main contributions

1. **Enhanced Tardos framework**:
- **Feasibility problem**: The paper proposes a method that can find a feasible solution or a Farkas certificate within \(O(m)\) calls to the approximate linear programming solver and \(O(mn^\omega)\) arithmetic operations.
- **Optimization problem**: For a feasible optimization problem, the method proposed in the paper can find the optimal solution pair within \(O(nm)\) calls to the approximate linear programming solver and \(O(mn^{\omega + 1})\) arithmetic operations.

2. **Utilize approximate solutions**:
- The paper effectively utilizes approximate solutions to construct exact solutions by combining the proximity theorems between approximate and exact solutions. This method avoids the high computational cost of directly using high - precision solvers.

3. **Condition number dependence**:
- The method proposed in the paper not only depends on the matrix condition number \(\bar{\chi}_A\) but can also convert the dependence from \(\bar{\chi}_A\) to a smaller \(\bar{\chi}_A^*\) through pre - processing, further improving the efficiency of the algorithm.

### Method overview

1. **Utilization of approximate solutions**:
- The paper utilizes approximate linear programming solvers (such as van den Brand's algorithm), obtaining approximate solutions by setting an appropriate precision \(\epsilon\). Then, these approximate solutions are transformed into exact solutions through the proximity theorem.

2. **Dynamic perturbation**:
- Different from the original Tardos framework, the method in the paper allows for dynamic adjustment of the perturbations of the objective function and the right - hand - side vector during the solving process, thus handling problems more flexibly.

3. **Recursive projection**:
- For infeasibility problems, the paper uses the recursive projection method to gradually remove variables and finally generate Farkas certificates. For optimization problems, the paper uses a similar method to gradually determine the support set of the optimal solution.

### Conclusion

By improving the Tardos framework and combining the latest advances in approximate solvers, the paper proposes an efficient and accurate linear programming solving method. This method is not only superior to existing methods in terms of computational complexity but can also handle infeasibility problems, having important theoretical and practical application values.