[FEA] Allow user to control solver method used at root node

[KyleBooth]

Describe the bug
I am using the Python API. My aim is to strictly use PDLP to solve the root node relaxation of a MIP problem. I am using the SolverMethod setting to set the method as PDLP, but the solver still uses barrier and simplex, which is unexpected.

Steps/Code to reproduce bug
Use:
settings.set_parameter("method", SolverMethod.PDLP)
In the Python API and invoke a solve. Solver log excerpt:

Solving LP root relaxation
Scaling matrix. Maximum column norm 3.044447e+00, minimum column norm 3.162278e-01
Dual Simplex Phase 1
Dual feasible solution found.
Dual Simplex Phase 2
 Iter     Objective           Num Inf.  Sum Inf.     Perturb  Time
    1 +3.1466918879784025e+04   98662 4.76112184e+09 0.00e+00 1.49
 1000 +3.1466918879784025e+04   98543 3.63115069e+09 0.00e+00 2.30
 2000 +3.1466918879784025e+04   98365 2.63039454e+09 0.00e+00 3.15
 3000 +3.1466918879784025e+04   98159 3.03303142e+10 0.00e+00 3.99
 4000 +3.1466918879784025e+04   98047 7.84284729e+11 5.34e-08 5.98
Running concurrent

PDLP finished
Status: Time Limit   Objective: 3.14669189e+04  Iterations: 1160  Time: 1.124s, Total time 1.170s
Barrier finished in 2.87 seconds
Concurrent time:  2.870s, total time 2.870s

Environment details (please complete the following information):
cuOpt version: 25.12.0
CPU: AMD EPYC 7R32, threads (physical/logical): 8/16, RAM: 37.89 GiB
CUDA 12.9, device: NVIDIA A10G (ID 0), VRAM: 21.99 GiB

[chris-maes]

There isn't currently a way to shut off dual simplex and barrier for the root relaxation. The concurrent mode runs all three methods.

[chris-maes]

We will take that as a feature request though :)

[KyleBooth]

OK thanks. Perhaps a tad better documentation for the "method" parameter as well then? What does it affect if it doesn't affect the root node LP solve? Does it only affect LP solves throughout the B&B tree?

[KyleBooth]

Additionally, if concurrent is being used above, I would expect it to terminate in 1.170s as this is when PDLP completes, however, it seems to wait for Barrier to complete before considering the root node LP "solved". Shouldn't concurrent terminate when the fastest method is done?

[shariharan2002]

Additionally, if concurrent is being used above, I would expect it to terminate in 1.170s as this is when PDLP completes, however, it seems to wait for Barrier to complete before considering the root node LP "solved". Shouldn't concurrent terminate when the fastest method is done?

I have the same question here. Thanks

[rgsl888prabhu]

@chris-maes for viz

[github-actions]

🔔 Hi @anandhkb @rg20, this issue has had no activity for 21 days. Please update or let us know if it can be closed. Thank you!

If this is an "epic" issue, then please add the "epic" label to this issue.
If it is a PR and not ready for review, then please convert this to draft.
If you just want to switch off this notification, then use the "skip inactivity reminder" label.

[github-actions]

🔔 Hi @anandhkb @rg20, this issue has had no activity for 21 days. Please update or let us know if it can be closed. Thank you!

If this is an "epic" issue, then please add the "epic" label to this issue.
If it is a PR and not ready for review, then please convert this to draft.
If you just want to switch off this notification, then use the "skip inactivity reminder" label.

[chris-maes]

Method controls the algorithm used to solve Linear Programs

We print finished in the logs when we have an LP solution. In branch and bound we need to have a basic solution. So we need to run crossover. We will improve the logging.