Support Sequential Backwards Pass

[dlcole3]

In the original version of PlasmoBenders.jl, the backwards pass was asynchronous, such that the cuts were all computed in parallel and added to the previous subproblems. The most common version of the backwards pass in literature though is to do a sequential solve, where the last problem is solved, cuts are passed back to the previous stage, that stage is solved, cuts are passed back, and so forth. This allows for information to propagate from the final stage to the first stage. This PR implements this option via the keyword argument sequential_backward_pass which is by default false.

[Copilot]

Pull Request Overview

This PR introduces a sequential_backward_pass option for the Benders decomposition algorithm, allowing cuts to be added during the backward pass itself rather than after. This represents an alternative execution strategy that may improve convergence in certain problem types.

• Added new sequential_backward_pass boolean option with corresponding infrastructure
• Refactored cut-adding logic into reusable functions _add_Benders_cut_to_object! and _add_strengthened_cut_to_object!
• Updated backward pass to iterate in reverse order and conditionally skip post-pass cut addition

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
lib/PlasmoBenders/src/Benders.jl	Added sequential_backward_pass option field, parameter, setter, and validation logic
lib/PlasmoBenders/src/solution.jl	Refactored cut-adding into separate functions and integrated sequential pass logic
lib/PlasmoBenders/src/utils.jl	Added sequential_backward_pass to boolean options list
lib/PlasmoBenders/src/PlasmoBenders.jl	Updated exports (with duplicate)
lib/PlasmoBenders/test/MIP_solve.jl	Added test cases for sequential backward pass
lib/PlasmoBenders/test/LP_solve.jl	Added test cases for sequential backward pass
lib/PlasmoBenders/examples/Benders_MILP.jl	Updated example to use new option
lib/PlasmoBenders/examples/Benders_MILP_parallel.jl	Updated example and cleaned up imports
lib/PlasmoBenders/Project.toml	Bumped version to 0.2.0

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This block is inside the for (j, object) in enumerate(next_objects) loop but should be outside it. The aggregated cut should only be added once after processing all objects, not once per object iteration.

Suggested change

	if !(get_multicut(optimizer))
	theta_vars = _get_theta(optimizer, last_object)
	theta_expr = sum(theta_vars[k] for k in 1:length(theta_vars))
	_add_cut_constraint!(optimizer, last_object, theta_expr, agg_rhs_expr)
	end
	end
	end
	if !(get_multicut(optimizer))
	theta_vars = _get_theta(optimizer, last_object)
	theta_expr = sum(theta_vars[k] for k in 1:length(theta_vars))
	_add_cut_constraint!(optimizer, last_object, theta_expr, agg_rhs_expr)
	end

[dlcole3]

No, I think this still needs to be inside the for loop because it needs to happen for each object.