Warm-start ignored in lexicographic optimization

[colonetti]

Hi, everyone!

First of, thanks for the amazing solver

I think I might have stumbled upon a little bug in HiGHS' lexicographic optimization. In all my experiments, whenever I add a linear objective function with addLinearObjective, HiGHS ignores the warm-start I provide.

Settings:

Python 3.11.7 highspy 1.11.0

Code to reproduce the behaviour:

import highspy as hp

m = hp.Highs()

x = m.addVariable(0, 1)
w = m.addVariable(0, 1)
y = m.addBinary()
z = m.addBinary()
m.addConstr(x == y, name="constr")
m.addConstr(w == z, name="constr_1")
m.addConstr(y + z == 1, name="constr_2")

m.setObjective(x + w)
m.setOptionValue("presolve", "off")
m.setSolution(4, [0, 1, 2, 3], [1, 0, 1, 0])
m.run()


print("\n\n\nNow with lexicography\n\n\n")


m = hp.Highs()

x = m.addVariable(0, 1)
w = m.addVariable(0, 1)
y = m.addBinary()
z = m.addBinary()
m.addConstr(x == y, name="constr")
m.addConstr(w == z, name="constr_1")
m.addConstr(y + z == 1, name="constr_2")

objs = [hp.HighsLinearObjective() for _ in range(2)]

for i, obj in enumerate(objs):
    obj.weight = 1
    obj.offset = 0
    obj.priority = i + 1
    obj.abs_tolerance = 0
    obj.rel_tolerance = 0
    if i == 0:
        obj.coefficients = [0, 0, 0, -1]
    else:
        obj.coefficients = [0, 0, -1, 0]
    m.addLinearObjective(obj)

m.setOptionValue("blend_multi_objectives", "off")
m.setOptionValue("presolve", "off")
m.setSolution(4, [0, 1, 2, 3], [1, 0, 1, 0])
m.run()

With a single objective function, added to the model with setObjective, HiGHS, as expected, evaluates the warm-start provided:

In the lexicographic optimization, it doesn't matter if I call the optimization using run, solve, or minimize. It also doesn't matter whether presolve and blend_multi_objectives are on or off. In any setting, HiGHS seems to simply ignore the solution provided.

Thanks in advance!

[jajhall]

OK thanks, I'll look at this

[jajhall]

Closed by #2450