Articles | Volume 18, issue 19
https://doi.org/10.5194/gmd-18-6951-2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.Special issue:
A bound-constrained formulation for complex solution phase minimization
Download
- Final revised paper (published on 09 Oct 2025)
- Supplement to the final revised paper
- Preprint (discussion started on 04 Nov 2024)
Interactive discussion
Status: closed
Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor
| : Report abuse
-
RC1: 'Comment on gmd-2024-197', Anonymous Referee #1, 15 Jan 2025
- AC1: 'Reply on RC1', Nicolas Riel, 27 Jan 2025
- CC1: 'Comment on gmd-2024-197', Paul Asimow, 20 May 2025
Peer review completion
AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload
AR by Nicolas Riel on behalf of the Authors (06 Jul 2025)
Author's response
Author's tracked changes
EF by Mario Ebel (09 Jul 2025)
Manuscript
ED: Referee Nomination & Report Request started (09 Jul 2025) by Mauro Cacace
RR by Anonymous Referee #2 (22 Jul 2025)

EF by Anna Glados (16 Jul 2025)
Supplement
ED: Publish subject to minor revisions (review by editor) (05 Aug 2025) by Mauro Cacace

AR by Nicolas Riel on behalf of the Authors (19 Aug 2025)
Author's response
Author's tracked changes
EF by Polina Shvedko (20 Aug 2025)
Manuscript
ED: Publish as is (01 Sep 2025) by Mauro Cacace

AR by Nicolas Riel on behalf of the Authors (02 Sep 2025)
The authors present a modification of their code MAGEMIN to allow for more efficient Gibbs free energy minimization. They state that the advantages of this modification arise from the new minimization algorithm being “unconstrained” as opposed to the “constrained” algorithm used previously. This distinction is misleading and the authors should change the paper to more clearly state the change in strategy. Simply put, the new algorithm is NOT “unconstrained”.
The central problem is simply stated: minimize the Gibbs free energy of a phase as a function of the fractions of cations X_i, I=1,n on one or more crystallographic sites. The minimization is subject to two constraints:
1) The cation fractions must sum to 1:
sum_i^n X_i = 1
and
2) they must be non-negative
X_i >= 0
Constraint (1) can be implemented either by performing a constrained minimization for which (1) forms an auxiliary statement to the minimization problem OR by minimizing over the null space of the constraint (1). The latter approach is appealing because it reduces the dimensionality of the problem (by the number of crystallographic sites), and avoids the auxiliary statement of constraint.
The primary focus of this paper is on the implementation of the null space approach. And it is on the basis of the removal of the auxiliary statement (1) that they describe their modified algorithm as “unconstrained”.
However, it is NOT unconstrained. The reason is constraint (2). This must still be dealt with: the minimization must still be subject to the bounds described by constraint 2. Thus the minimization is still constrained, not unconstrained. In other words, they have removed an equality constraint, but NOT the inequality constraint.
Of course the authors DO end up applying the inequality constraint, although this application is somewhat buried in the details (Eq. 32).
My argument is not with the method itself, nor primarily with the claims of greater efficiency (although see below). It is with what I think is a misleading description of the modification. I implore them to characterize their modification more clearly and correctly.
Some further comments.
1) Readers may get the mistaken impression that the null space approach is new in the context of petrological Gibbs free energy minimization codes. It is not, and an example of another minimization code that uses the null space approach is HeFESTo (Stixrude and Lithgow-Bertelloni, 2011). Nor do the authors of HeFESTo claim priority for the null space approach, but it is the closest implementation known to this reviewer in terms of intended application (i.e. to petrology).
2) The claimed superiority of their new implementation to SLSQP is puzzling to this reviewer. One reason for the source of puzzlement is that HeFESTo uses SLSQP and finds a very high (essentially perfect) rate of success and precision as documented in Stixrude and Lithgow-Bertelloni (2021). Perhaps the reason is that HeFESTo uses the null space approach for the equality constraints and relies on the constraint facility of SLSQP only for the inequality constraint. Maybe in the current paper, the authors are instead relying on SLSQP to take care of the equality AND inequality constraints. Therefore, to avoid confusion in the literature, I suggest the following test:
Perform the SLSQP minimization(s) again, but by combining SLSQP with the null space approach.
3) A final comment on SLSQP. The authors state that SLSQP sometimes fails because it violates inequality constraints and then cannot return to the feasible space. But this should not be true. According to NLOPT documentation SLSQP is guaranteed to respect inequality constraints at all intermediate steps of the minimization. Perhaps the problem is that SLSQP sometimes does venture into the space where one or more X_i are exactly zero. If this is the case, it is a problem easily solved, by setting the inequality constraint instead to:
X_i >= epsilon
very much like their own implementation of the inequality constraint.