[Minuit2] Make function passing purely std::function based

[guitargeek]

This makes is possible to use Minuit2 directly from Python, as we can benefit from the std::function pythonization.

It also makes the Minuit2 code more consistent, because std::function was already used for the G2 (second derivatives) and the Hessian.

Here is an example of how the base Minuit 2 can be used from Python now:

import ROOT

# Define function to minimize
def fcn(params):
    x, y = params[0], params[1]
    return (x - 2.0)**2 + (y + 1.0)**2 + 1.0

# Wrap into Minuit2 FCNBase
func = ROOT.Minuit2.FCNAdapter(fcn)

# Initial values and step sizes as std::vector<double>
start_vals = ROOT.std.vector['double']([0.0, 0.0])
step_sizes = ROOT.std.vector['double']([0.1, 0.1])

# Create the minimizer
minimizer = ROOT.Minuit2.VariableMetricMinimizer()


state = ROOT.Minuit2.MnUserParameterState(start_vals, step_sizes)

# Run minimization directly
result = minimizer.Minimize(func, state)

# Inspect results
print("Minimization valid?", result.IsValid())
print("Minimum f =", result.Fval())
print("Best-fit parameters:")
for i in range(result.UserParameters().Params().size()):
    val = result.UserParameters().Params()[i]
    err = result.UserParameters().Errors()[i]
    print(f"  param[{i}] = {val} ± {err}")

This makes the base Minuit 2 code easily usable from Python, as promised in the ROOT plan of work. With "base Minuit 2 code", I mean the same classes that would also be built with Minuit 2 standalone, not using any other ROOT abstraction layers like the ROOT::Math::Minimizer interface, the ROOT::Math::Fitter or RooFit.

[hageboeck]

LGTM, but note that an entry in the release notes might be in order:

• FCNGradAdapter.h is gone, so code might fail to compile
• FCNAdapter<xxx> has to be converted to FCNAdapter

[github-actions]

Test Results

    22 files      22 suites   3d 20h 39m 24s ⏱️
 3 686 tests  3 686 ✅ 0 💤 0 ❌
79 196 runs  79 196 ✅ 0 💤 0 ❌

Results for commit 22b91bb.

♻️ This comment has been updated with latest results.