PLS (complexity)
In computational complexity theory, Polynomial Local Search is a complexity class that models the difficulty of finding a locally optimal solution to an optimization problem. The main characteristics of problems that lie in PLS are that the cost of a solution can be calculated in polynomial time and the neighborhood of a solution can be searched in polynomial time. Therefore it is possible to verify whether or not a solution is a local optimum in polynomial time.
Furthermore, depending on the problem and the algorithm that is used for solving the problem, it might be faster to find a local optimum instead of a global optimum.
Description
When searching for a local optimum, there are two interesting issues to deal with: First how to find a local optimum, and second how long it takes to find a local optimum. For many local search algorithms, it is not known, whether they can find a local optimum in polynomial time or not. So to answer the question of how long it takes to find a local optimum, Johnson, Papadimitriou and Yannakakis introduced the complexity class PLS in their paper "How easy is local search". It contains local search problems for which the local optimality can be verified in polynomial time.A local search problem is in PLS, if the following properties are satisfied:
- The size of every solution is polynomially bounded in the size of the instance.
- It is possible to find some solution of a problem instance in polynomial time.
- It is possible to calculate the cost of each solution in polynomial time.
- It is possible to find all neighbors of each solution in polynomial time.
Example
Consider the following instance of the Max-2Sat Problem:. The aim is to find an assignment, that maximizes the sum of the satisfied clauses.A solution for that instance is a bit string that assigns every the value 0 or 1. In this case, a solution consists of 3 bits, for example, which stands for the assignment of to with the value 0. The set of solutions is the set of all possible assignments of, and.
The cost of each solution is the number of satisfied clauses, so because the second and third clause are satisfied.
The Flip-neighbor of a solution is reached by flipping one bit of the bit string, so the neighbors of are with the following costs:
There are no neighbors with better costs than, if we are looking for a solution with maximum cost. Even though is not a global optimum, is a local optimum, because none of its neighbors has better costs.
Intuitively it can be argued that this problem lies in PLS, because:
- It is possible to find a solution to an instance in polynomial time, for example by setting all bits to 0.
- It is possible to calculate the cost of a solution in polynomial time, by going once through the whole instance and counting the clauses that are satisfied.
- It is possible to find all neighbors of a solution in polynomial time, by taking the set of solutions that differ from in exactly one bit.
Formal Definition
- The size of every solution is polynomial bounded in the size of
- Problem instances and solutions are polynomial time verifiable
- There is a polynomial time computable function that returns for each instance some solution
- There is a polynomial time computable function that returns for each solution of an instance the cost
- There is a polynomial time computable function that returns the set of neighbors for an instance-solution pair
- There is a polynomial time computable function that returns a neighboring solution with better cost than solution, or states that is locally optimal
- For every instance, exactly contains the pairs where is a local optimal solution of
A local optimum is a solution, that has no neighbor with better costs. In the implicit graph, a local optimum is a sink. A neighborhood where every local optimum is a global optimum, which is a solution with the best possible cost, is called an exact neighborhood.
Example neighborhood structures
Example neighborhood structures for problems with boolean variables as solution:- Flip - The neighborhood of a solution can be achieved by negating one arbitrary input bit. So one solution and all its neighbors have Hamming distance one:.
- Kernighan-Lin - A solution is a neighbor of solution if can be obtained from by a sequence of greedy flips, where no bit is flipped twice. This means, starting with, the Flip-neighbor of with the best cost, or the least loss of cost, is chosen to be a neighbor of s in the Kernighan-Lin structure. As well as best neighbor of , and so on, until is a solution where every bit of is negated. Note that it is not allowed to flip a bit back, if it once has been flipped.
- k-Flip - A solution is a neighbor of solution if the Hamming distance between and is at most, so.
- Swap - A partition of nodes in a graph is a neighbor of a partition if can be obtained from by swapping one node with a node.
- Kernighan-Lin - A partition is a neighbor of if can be obtained by a greedy sequence of swaps from nodes in with nodes in. This means the two nodes and are swapped, where the partition gains the highest possible weight, or loses the least possible weight. Note that no node is allowed to be swapped twice.
- Fiduccia-Matheyses - This neighborhood is similar to the Kernighan-Lin neighborhood structure, it is a greedy sequence of swaps, except that each swap happens in two steps. First the with the most gain of cost, or the least loss of cost, is swapped to, then the node with the most cost, or the least loss of cost is swapped to to balance the partitions again. Experiments have shown that Fiduccia-Mattheyses has a smaller run time in each iteration of the standard algorithm, though it sometimes finds an inferior local optimum.
- FM-Swap - This neighborhood structure is based on the Fiduccia-Mattheyses neighborhood structure. Each solution has only one neighbor, the partition obtained after the first swap of the Fiduccia-Mattheyses.
The standard Algorithm
Given some instance of a PLS problem, find a locally optimal solution such that for all .
Every local search problem can be solved using the following iterative improvement algorithm:
- Use to find an initial solution
- Use algorithm to find a better solution. If such a solution exists, replace by and repeat step 2, else return
The run time of the standard algorithm is pseudo-polynomial in the number of different costs of a solution.
The space the standard algorithm needs is only polynomial. It only needs to save the current solution, which is polynomial bounded by definition.
Reductions
A Reduction of one problem to another may be used to show that the second problem is at least as difficult as the first. In particular, a PLS-reduction is used to prove that a local search problem that lies in PLS is also PLS-complete, by reducing a PLS-complete Problem to the one that shall be proven to be PLS-complete.PLS-reduction
A local search problem is PLS-reducable to a local search problem if there are two polynomial time functions and such that:- if is an instance of , then is an instance of
- if is a solution for of , then is a solution for of
- if is a local optimum for instance of , then has to be a local optimum for instance of
PLS-reductions are transitive.
Tight PLS-reduction
Definition [|Transition graph]
The transition graph of an instance of a problem is a directed graph. The nodes represent all elements of the finite set of solutions and the edges point from one solution to the neighbor with strictly better cost. Therefore it is an acyclic graph. A sink, which is a node with no outgoing edges, is a local optimum.The height of a vertex is the length of the shortest path from to the nearest sink.
The height of the transition graph is the largest of the heights of all vertices, so it is the height of the largest shortest possible path from a node to its nearest sink.
Definition Tight PLS-reduction
A PLS-reduction from a local search problem to a local search problem is atight PLS-reduction if for any instance of, a subset of solutions
of instance of can be chosen, so that the following properties are satisfied:
- contains, among other solutions, all local optima of
- For every solution of , a solution of can be constructed in polynomial time, so that
- If the transition graph of contains a direct path from to, and, but all internal path vertices are outside, then for the corresponding solutions and holds either or contains an edge from to
Relationship to other complexity classes
PLS also is a subclass of TFNP, that describes computational problems in which a solution is guaranteed to exist and can be recognized in polynomial time. For a problem in PLS, a solution is guaranteed to exist because the minimum-cost vertex of the entire graph is a valid solution, and the validity of a solution can be checked by computing its neighbors and comparing the costs of each one to another.
It is also proven that if a PLS problem is NP-hard, then NP = co-NP.
PLS-completeness
Definition
A local search problem is PLS-complete, if- is in PLS
- every problem in PLS can be PLS-reduced to
List of PLS-complete Problems
This is an incompete list of some known problems that are PLS-complete.Notation: Problem / Neighborhood structure
- Min/Max-circuit/Flip has been proven to be the first PLS-complete problem.
- Positive-not-all-equal-max-3Sat/Flip has been proven to be PLS-complete via a tight PLS-reduction from Min/Max-circuit/Flip to Positive-not-all-equal-max-3Sat/Flip. Note that Positive-not-all-equal-max-3Sat/Flip can be reduced from Max-Cut/Flip too.
- Positive-not-all-equal-max-3Sat/Kernighan-Lin has been proven to be PLS-complete via a tight PLS-reduction from Min/Max-circuit/Flip to Positive-not-all-equal-max-3Sat/Kernighan-Lin.
- Max-2Sat/Flip has been proven to be PLS-complete via a tight PLS-reduction from Max-Cut/Flip to Max-2Sat/Flip.
- Min-4Sat-B/Flip has been proven to be PLS-complete via a tight PLS-reduction from Min-circuit/Flip to Min-4Sat-B/Flip.
- Max-4Sat-B/Flip has been proven to be PLS-complete via a PLS-reduction from Max-circuit/Flip to Max-4Sat-B/Flip.
- Max-4Sat-/Flip has been proven to be PLS-complete via a PLS-reduction from Max-circuit/Flip to Max-4Sat-/Flip.
- Max-Uniform-Graph-Partitioning/Swap has been proven to be PLS-complete via a tight PLS-reduction from Max-Cut/Flip to Max-Uniform-Graph-partitioning/Swap.
- Max-Uniform-Graph-Partitioning/Fiduccia-Matheyses is stated to be PLS-complete without proof.
- Max-Uniform-Graph-Partitioning/FM-Swap has been proven to be PLS-complete via a tight PLS-reduction from Max-Cut/Flip to Max-Uniform-Graph-partitioning/FM-Swap.
- Max-Uniform-Graph-Partitioning/Kernighan-Lin has been proven to be PLS-complete via a PLS-reduction from Min/Max-circuit/Flip to Max-Uniform-Graph-Partitioning/Kernighan-Lin. There is also a tight PLS-reduction from Positive-not-all-equal-max-3Sat/Kernighan-Lin to Max-Uniform-Graph-Partitioning/Kernighan-Lin.
- Max-Cut/Flip has been proven to be PLS-complete via a tight PLS-reduction from Positive-not-all-equal-max-3Sat/Flip to Max-Cut/Flip.
- Max-Cut/Kernighan-Lin is claimed to be PLS-complete without proof.
- Min-Independent-Dominating-Set-B/k-Flip has been proven to be PLS-complete via a tight PLS-reduction from Min-4Sat-B’/Flip to Min-Independent-Dominating-Set-B/k-Flip.
- Weighted-Independent-Set/Change is claimed to be PLS-complete without proof.
- Maximum-Weighted-Subgraph-with-property-P/Change is PLS-complete if property P = ”has no edges”, as it then equals Weighted-Independent-Set/Change. It has also been proven to be PLS-complete for a general hereditary, non-trivial property P via a tight PLS-reduction from Weighted-Independent-Set/Change to Maximum-Weighted-Subgraph-with-property-P/Change.
- Set-Cover/k-change has been proven to be PLS-complete for each k ≥ 2 via a tight PLS-reduction from -Max-Constraint-Assignment/Change to Set-Cover/k-change.
- Metric-TSP/k-Change has been proven to be PLS-complete via a PLS-reduction from Max-4Sat-B/Flip to Metric-TSP/k-Change.
- Metric-TSP/Lin-Kernighan has been proven to be PLS-complete via a tight PLS-reduction from Max-2Sat/Flip to Metric-TSP/Lin-Kernighan.
- Local-Multi-Processor-Scheduling/k-change has been proven to be PLS-complete via a tight PLS-reduction from Weighted-3Dimensional-Matching/-Swap to Local-Multi-Processor-scheduling/-change, where ≥ 8.
- Selfish-Multi-Processor-Scheduling/k-change-with-property-t has been proven to be PLS-complete via a tight PLS-reduction from Weighted-3Dimensional-Matching/-Swap to -Selfish-Multi-Processor-Scheduling/k-change-with-property-t, where ≥ 8.
- Finding the pure Nash Equilibrium in a General-Congestion-Game/Change has been proven PLS-complete via a tight PLS-reduction from Positive-not-all-equal-max-3Sat/Flip to General-Congestion-Game/Change.
- Finding the pure Nash Equilibrium in a Symmetric General-Congestion-Game/Change has been proven to be PLS-complete via a tight PLS-reduction from an asymmetric General-Congestion-Game/Change to symmetric General-Congestion-Game/Change.
- Finding a pure pure Nash Equilibrium in an Asymmetric Directed-Network-Congestion-Games/Change has been proven to be PLS-complete via a tight reduction from Positive-not-all-equal-max-3Sat/Flip to Directed-Network-Congestion-Games/Change and also via a tight PLS-reduction from 2-Threshold-Games/Change to Directed-Network-Congestion-Games/Change.
- Finding a pure pure Nash Equilibrium in an Asymmetric Undirected-Network-Congestion-Games/Change has been proven to be PLS-complete via a tight PLS-reduction from 2-Threshold-Games/Change to Asymmetric Undirected-Network-Congestion-Games/Change.
- Finding a pure pure Nash Equilibrium in a 2-Threshold-Game/Change has been proven to be PLS-complete via a tight reduction from Max-Cut/Flip to 2-Threshold-Game/Change.
- Finding a pure Nash Equilibrium in Market-Sharing-Game/Change with polynomial bounded costs has been proven to be PLS-complete via a tight PLS-reduction from 2-Threshold-Games/Change to Market-Sharing-Game/Change.
- Finding a pure Nash Equilibrium in an Overlay-Network-Design/Change has been proven to be PLS-complete via a reduction from 2-Threshold-Games/Change to Overlay-Network-Design/Change. Analogously to the proof of asymmetric Directed-Network-Congestion-Game/Change, the reduction is tight.
- Min-0-1-Integer Programming/k-Flip has been proven to be PLS-complete via a tight PLS-reduction from Min-4Sat-B’/Flip to Min-0-1-Integer Programming/k-Flip.
- Max-0-1-Integer Programming/k-Flip is claimed to be PLS-complete because of PLS-reduction to Max-0-1-Integer Programming/k-Flip, but the proof is left out.
- -Max-Constraint-Assignment
- * -Max-Constraint-Assignment-3-partite/Change has been proven to be PLS-complete via a tight PLS-reduction from Circuit/Flip to -Max-Constraint-Assignment-3-partite/Change.
- * -Max-Constraint-Assignment-2-partite/Change has been proven to be PLS-complete via a tight PLS-reduction from Circuit/Flip to -Max-Constraint-Assignment-2-partite/Change.
- * -Max-Constraint-Assignment/Change has been proven to be PLS-complete via a tight reduction from Circuit/Flip to -Max-Constraint-Assignment/Change.
- * -Max-Constraint-Assignment/Change equals Max-4Sat-/Flip and has been proven to be PLS-complete via a PLS-reduction from Max-circuit/Flip. It is claimed that the reduction can be extended so tightness is obtained.
- Nearest-Colorful-Polytope/Change has been proven to be PLS-complete via a PLS-reduction from Max-2Sat/Flip to Nearest-Colorful-Polytope/Change.
- Stable-Configuration/Flip in a Hopfield network has been proven to be PLS-complete if the thresholds are 0 and the weights are negative via a tight PLS-reduction from Max-Cut/Flip to Stable-Configuration/Flip.
- Weighted-3Dimensional-Matching/-Swap has been proven to be PLS-complete for p ≥9 and q ≥ 15 via a tight PLS-reduction from -Max-Constraint-Assignment-2-partite/Change to Weighted-3Dimensional-Matching/-Swap.