Limitations of the Simple Rule

Oct 13, 2025 by sorusgentrification.com 56 views

Introduction

Warm starting is a crucial feature in linear programming solvers that allows the solver to leverage the previous solution to improve the efficiency of the current solution. However, the simple rule of "hot start the unreduced LP if there is a basis" can be far from optimal in many scenarios. In this article, we will explore the limitations of this rule and propose improvements to warm starting after model changes.

Limitations of the Simple Rule

The simple rule of "hot start the unreduced LP if there is a basis" assumes that the basis is still valid and that the steepest edge weights can be computed efficiently. However, this assumption is not always true. Here are some scenarios where the simple rule may not be optimal:

Basis is No Longer Square

When the basis is no longer square, the steepest edge weights cannot be computed efficiently, and the solver may need to perform additional computations to update the basis. In this case, the simple rule may not be optimal, and the solver may need to perform a full restart.

Steepest Edge Weights Have to be Computed

If the steepest edge weights have to be computed, the solver may need to perform additional computations to update the basis. This can be time-consuming, especially if the number of variables is large.

Presolve Reduces the LP Greatly

If the presolve reduces the LP greatly, the solver may need to perform a full restart to take advantage of the reduced size of the LP. In this case, the simple rule may not be optimal, and the solver may need to perform a full restart.

Many Modifications Have Been Made

If many modifications have been made to the model, the solver may need to perform a full restart to ensure that the solution is optimal. In this case, the simple rule may not be optimal, and the solver may need to perform a full restart.

Improving Warm Starting

To improve warm starting after model changes, we propose the following:

Clear Cycling Detection

When re-starting, switching simplex algorithms, or removing perturbations, the cycling detection should be cleared to avoid false positives. This can be achieved by setting the cycling detection flag to zero and resetting the cycling detection variables.

Report Taboo Basis Change

If a taboo basis change triggers model status kUnknown, the solver should report what that basis change it was. This can help the user to understand what went wrong and how to fix it.

Tie to #1979

The cycling detection should be tied to #1979 to ensure that the cycling detection is cleared when re-starting, switching simplex algorithms, or removing perturbations.

Implementation

To implement the improvements proposed above, the following changes can be made to the solver:

Clear Cycling Detection

The cycling detection flag should be set to zero and the cycling detection variables should be reset when re-starting, switching simplex algorithms, or removing perturbations.

Report Taboo Basis Change

The solver should report what basis change triggered the model status kUnknown when a taboo basis change occurs.

Tie to #1979

The cycling detection should be tied to #1979 to ensure that the cycling detection is cleared when re-starting, switching simplex algorithms, or removing perturbations.

Conclusion

Warm starting is a crucial feature in linear programming solvers that allows the solver to leverage the previous solution to improve the efficiency of the current solution. However, the simple rule of "hot start the unreduced LP if there is a basis" can be far from optimal in many scenarios. By clearing cycling detection, reporting taboo basis changes, and tying to #1979, we can improve warm starting after model changes and make the solver more efficient.

Future Work

Future work can include:

Improving Steepest Edge Weights Computation

The steepest edge weights computation can be improved by using more efficient algorithms or by parallelizing the computation.

Presolve Reduction

The presolve reduction can be improved by using more efficient algorithms or by parallelizing the computation.

Model Modification Handling

The model modification handling can be improved by using more efficient algorithms or by parallelizing the computation.

References

[1] "Warm Starting in Linear Programming Solvers" by [Author]
[2] "Improving Warm Starting after Model Changes" by [Author]

Appendix

The appendix includes the implementation details of the proposed improvements.

Clear Cycling Detection

The cycling detection flag is set to zero and the cycling detection variables are reset when re-starting, switching simplex algorithms, or removing perturbations.

Report Taboo Basis Change

The solver reports what basis change triggered the model status kUnknown when a taboo basis change occurs.

Tie to #1979

Introduction

Q&A

Q: What is warm starting in linear programming solvers?

A: Warm starting is a feature in linear programming solvers that allows the solver to leverage the previous solution to improve the efficiency of the current solution.

Q: What is the simple rule of warm starting?

A: The simple rule of warm starting is to "hot start the unreduced LP if there is a basis".

Q: What are the limitations of the simple rule?

A: The simple rule assumes that the basis is still valid and that the steepest edge weights can be computed efficiently. However, this assumption is not always true. The basis may be no longer square, the steepest edge weights may have to be computed, the presolve may reduce the LP greatly, or many modifications may have been made to the model.

Q: How can we improve warm starting after model changes?

A: We can improve warm starting after model changes by clearing cycling detection, reporting taboo basis changes, and tying to #1979.

Q: What is cycling detection?

A: Cycling detection is a feature in linear programming solvers that detects when the solver is cycling, i.e., when the solver is making no progress.

Q: Why is it important to clear cycling detection?

A: It is important to clear cycling detection to avoid false positives. When the solver is re-started, switched to a different simplex algorithm, or has perturbations removed, the cycling detection should be cleared to ensure that the solver is not stuck in a cycle.

Q: What is a taboo basis change?

A: A taboo basis change is a change to the basis that is not allowed by the solver. This can happen when the solver is trying to solve a linear program with a non-convex objective function.

Q: Why is it important to report taboo basis changes?

A: It is important to report taboo basis changes to the user so that they can understand what went wrong and how to fix it.

Q: What is #1979?

A: #1979 is a reference to a specific issue in the solver that is related to cycling detection.

Q: Why is it important to tie cycling detection to #1979?

A: It is important to tie cycling detection to #1979 to ensure that the cycling detection is cleared when re-starting, switching simplex algorithms, or removing perturbations.

Conclusion

Warm starting is a crucial feature in linear programming solvers that allows the solver to leverage the previous solution to improve the efficiency of the current solution. However, the simple rule of "hot start the unreduced LP if there is a basis" can be far from optimal in many scenarios. By clearing cycling detection, reporting taboo basis changes, and tying to #1979, we can improve warm starting after model changes and make the solver more efficient.

Future Work

Future work can include:

Improving Steepest Edge Weights Computation

The steepest edge weights computation can be improved by using more efficient algorithms or by parallelizing the computation.

Presolve Reduction

The presolve reduction can be improved by using more efficient algorithms or by parallelizing the computation.

Model Modification Handling

The model modification handling can be improved by using more efficient algorithms or by parallelizing the computation.

References

[1] "Warm Starting in Linear Programming Solvers" by [Author]
[2] "Improving Warm Starting after Model Changes" by [Author]

Appendix

The appendix includes the implementation details of the proposed improvements.

Clear Cycling Detection

The cycling detection flag is set to zero and the cycling detection variables are reset when re-starting, switching simplex algorithms, or removing perturbations.

Report Taboo Basis Change

The solver reports what basis change triggered the model status kUnknown when a taboo basis change occurs.

Tie to #1979

The cycling detection is tied to #1979 to ensure that the cycling detection is cleared when re-starting, switching simplex algorithms, or removing perturbations.