Jesús A. De Loera,Raymond Hemmecke,Matthias Köppe,Robert Weismantel
Abstract: We classify, according to their computational complexity, integer optimization problems whose constraints and objective functions are polynomials with integer coefficients and the number of variables is fixed. For the optimization of an integer polynomial over the lattice points of a convex polytope, we show an algorithm to compute lower and upper bounds for the optimal value. For polynomials that are non-negative over the polytope, these sequences of bounds lead to a fully polynomial-time approximation scheme for the optimization problem.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the computational complexity classification of integer polynomial optimization in a fixed dimension. Specifically, the authors study integer optimization problems where both the objective function and the constraint conditions are polynomials with integer coefficients when the number of variables is fixed. For the problem of optimizing an integer polynomial on the lattice points of a convex polyhedron, the authors propose an algorithm to calculate the upper and lower bounds of the optimal value. If the polynomial is non - negative on the polyhedron, these bounds can form a fully polynomial - time approximation scheme (FPTAS) to solve the optimization problem.
### Main Contributions
1. **Computational Complexity Classification**:
- The paper classifies the computational complexity of different types of polynomial optimization problems, including optimization problems under linear constraints, convex semi - algebraic constraints, and arbitrary polynomial constraints. These results are summarized in Table 1, with new results marked by letters, known results marked by asterisks, and arrows indicating implication relationships.
2. **FPTAS for Non - negative Polynomials**:
- For the problem of optimizing an arbitrary integer polynomial on the lattice points of a convex rational polyhedron in a fixed dimension, the authors propose an algorithm to generate the upper and lower bounds of the optimal value. In particular, if the polynomial is non - negative on the entire polyhedron, the algorithm can provide a fully polynomial - time approximation scheme (FPTAS).
### Key Technologies
- **Barvinok's Theory**: Using Barvinok's theory, all lattice points of a polyhedron can be encoded by rational functions. This enables the generating function of the values of the objective function on all lattice points to be calculated in polynomial time.
- **Differential Operators**: By applying differential operators related to the objective function, the rational function representation of the generating function can be calculated, and further the sum of the values of the objective function on all lattice points can be calculated.
### Algorithm Steps
1. **Pre - processing**: If the objective function is not entirely non - negative on the polyhedron, find the upper and lower bounds of the variables by solving a linear program and convert the objective function into a non - negative polynomial.
2. **Generating Function Calculation**: Use Barvinok's algorithm to calculate the generating function of the polyhedron lattice points and calculate the number of lattice points from the generating function.
3. **Upper and Lower Bound Calculation**: By applying Lemma 3.1, calculate the generating function of the values of the objective function on all lattice points and extract the upper and lower bounds from it.
4. **Termination Condition**: When the difference between the upper and lower bounds is less than 1, return the integer part of the upper bound as the optimal value.
### Application Expansion
- **Mixed - integer Case**: The paper also discusses how to extend the above results to the mixed - integer case, that is, the situation where some variables are allowed to be continuous. By constructing a series of sub - problems, the optimal value of the mixed - integer optimization problem can be approximated.
### Conclusion
Through detailed research and algorithm design, this paper provides an important theoretical basis and practical algorithm tools for integer polynomial optimization problems in a fixed dimension.