Fix issue when AMPGO method is aborted due to reaching max_nfev

[dangarciahe]

Description

This PR adds a fix to the AMPGO method for the minize() function. Previously, when AMPGO is aborted due to reaching max_nfev, it will return the last evaluated values, instead of the best value found so far, an incorrect behavior per AMPGO paper. This was discussed in #1017. The new behavior fixes this, and will return the best value found so far, even if the process is aborted due to max_nfev.

Additionally, a new test is added: It verifies against a known case to exhibit this troubled behavior in the current version, but returns a correct value under the newer version proposed by this PR.

This should improve AMPGO method, specially when minimizing computationally slow functions where an arbitrarily big max_nfev is not feasible, and a test to prevent future code to accidentally re-introduce this behavior.

Type of Changes

• Bug fix
• New feature
• Refactoring / maintenance
• Documentation / examples

Tested on

Python: 3.14.0 | packaged by Anaconda, Inc. | (main, Oct 8 2025, 17:15:28) [GCC 11.2.0]

lmfit: 1.3.4, scipy: 1.16.2, numpy: 2.3.4, asteval: 1.0.6, uncertainties: 3.2.3

Verification

Have you

• included docstrings that follow PEP 257?
• referenced existing Issue and/or provided relevant link to mailing list?
• verified that existing tests pass locally?
• verified that the documentation builds locally?
• squashed/minimized your commits and written descriptive commit messages?
• added or updated existing tests to cover the changes?
• updated the documentation and/or added an entry to the release notes (doc/whatsnew.rst)?
• added an example?

Minimal reproducible example

Here's a minimal example that allows to see the improvements:

import numpy as np
import lmfit

np.random.seed(10)

def parabolic(p):
    x = float(p["x"])
    y = x**2
    print(f"x: {x:.5e}, y:{y:.5e}")
    return y

fit_params = lmfit.Parameters()
fit_params.add('x', value=-2, min=-10, max=10, vary=True)
result = lmfit.minimize(parabolic, fit_params, method="ampgo", max_nfev=100)

print(lmfit.fit_report(result))

Under the older version, the fit will return x: 0.30768323 despite the existence of several evaluations of x in the order of x=-4.99985e-08, as evidenced by the prints; new version will correctly return x: -4.9999e-08

[reneeotten]

thansk @dangarciahe for pointing out the issue and providing a solution! And, yes, we should certainly fix this.

I am just not sure whether that should be done in the actual _ampgo.py implementation or whether it can be done somehow in the ampgo finction minimizer.py. The latter would be preferable - I'll need to look.

[newville]

@dangarciahe @reneeotten Thanks! I would probably have a slight preference for doing this in minimizer.py, but wouldn't be adamant about that.

[dangarciahe]

Hi @newville @reneeotten. I would be happy to make those changes, but have no idea on how to, given that, from what I underestand, if _ampgo.py aborts, minizer.py has no idea of the current best value, so I don't see how could I catch those values. Hence why I added the try-except block to _ampgo.

Some guidance in this would be appreciated.

[newville]

please have AbortFitException set ret = None
and then make the test here be if ret is not None.

[newville]

result.nfev -= 2 if result.aborted else 1

might be worth considering here

[newville]

        except Exception as exc:
            if isinstance(exc, AbortFitException):
              ...

[newville]

.... or bettter....

        except AbortFitException:
               do things....
        except Exception as exc: 
              do other things... 

[newville]

@dangarciahe Thanks - I think a few small changes and this will be a great addition. Happy to hear others thoughts