Hi,
I need to do a search for materials with ferromagnetic ordering. Is it possible to add this as one of the search parameters?
Thanks in advance,
BS
Hi @Shortall_Brian,
Itâs possible to query on magnetic_type, e.g. use {'magnetic_type':'FM'} in the materials explorer search or via an MP API query in pymatgen.
Note that this is not a prediction that a given material is ferromagnetic, but rather gives calculations where a material has been calculated with a ferromagnetic ordering.
Best,
Matt
1 Like
Hi @mkhorton,
Just to clarify, is this to say that the lowest energy magnetic configuration for this material is ferromagnetic, or that the calculations were performed with ferromagnetic ordering? I realize theyâre not mutually exclusive, but Iâm just looking to clarify the wording of your response a bit!
Best,
Vanessa
Hi @Vanessa_Meschke, welcome!
Yes, youâre correct that these are not mutually exclusive.
Historically, materials in Materials Project were only been calculated with a ferromagnetic order initialized by default (some of these will relax to a non-FM order).
However, about a year or so ago we started explicitly calculating multiple magnetic orderings for a given material (see method here). We have now predicted magnetic orderings for about 5000 materials, primarily transition metal oxides, with more being calculated on a continual basis.
There currently isnât an easy way to distinguish these two cases with the API or via search. Weâll be added an additional key to make explicitly clear which materials we have these predictions for. For now, itâs a fair assumption that if you see a magnetic material that has a lot of associated calculations with it (see the footer of the material details page), it probably means itâs gone through this workflow.
Hope this helps! Happy to answer additional questions.
Best,
Matt
Perfect, thanks so much Matt! That does clear it up!
Hi @mkhorton,
I wanted to check back in on this topic. What are currently the best practices for pulling:
- calculations of a mp-id with a specific magnetic order
- all magnetic calculations (varying initial conditions) for a given mp-id
- the lowest energy magnetic calculation for a given mp-id
Thanks!
Tess
Hi @blondegeek,
By default all that Materials Project currently offers is 3. â the lowest energy magnetic calculation for a given mp-id.
You can retrieve 1. and 2. manually by first requesting the task_ids for a given mp-id (the full list of calculations associated with that material), and then retrieving the structure associated with each task.
I intend to add a more intuitive way to explore this data on our new API and website but have not done so yet (sign-up form available here to help us test for when we send out an announcement email with more info).
In the meantime, the data associated from our recent paper by @ncfrey is available here, this is a database dump of a few thousand transition metal oxides, and this database dump comes directly from the magnetic_orderings collection we generate from atomate, so is a lot easier to query for the kind of information you need. This is (estimate) 90% of the magnetic ordering information available on MP right now, so might be an easier place to start without missing much.
Best,
Matt
Hi @mkhorton
I have queried the MP database based on chemical classes of interest, which resulted in a couple of thousands of materials. I am interested in the magnetic ordering of these materials for the purpose of developing a ML classifier, and I have extracted this information (which is under the header âMagnetic Orderingâ) via web scrapping.
After reading through this tread, I am confused about what this magnetic ordering actually indicates. My understanding is that the DFT calculations in MP uses ferromagnetic initialization by default. During the DFT relaxation of a unit-cell, there will be convergence of parameters, including the spin state, which can lead to non-FM states. And the information under âMagnetic Orderingâ is the magnetic ordering of a material that has gone through geometrical relaxation. However, your post from 2018 (I think) is implying that this information is the initialization strategy used in the calculation before the relaxation, rather that the magnetic ordering after the relaxation. Can you clarify this please?
If my understanding is incorrect and the information under âMagnetic Orderingâ indicates the initialization strategy before geometrical relaxation, what is the correct way to think when I have materials in my data collection where the magnetic ordering is stated as an AFM? Is this where the AFM initialization strategy was used? Or is this when the FM initialization strategy was used and the magnetic ordering changed to non-FM (i.e. AFM in this case) after the relaxation? I am trying to understand how I should interpret the magnetic ordering labels I have web scrapped from the MP website, especially if it states that it is of an AFM (or NM) order.
Many thanks,
Son
Hi @sgj30,
Thanks for your interest, Iâm glad this data may be helpful for you. There are a few points I might help clear up:
I have extracted this information (which is under the header âMagnetic Orderingâ) via web scrapping.
Web scraping should never be necessary for Materials Project data. We provide a free and open API for retrieving information. This is helpful for both yourself since this will allow faster retrieval of information, and also helpful for us because it is kinder on our web servers.
After reading through this tread, I am confused about what this magnetic ordering actually indicates. My understanding is that the DFT calculations in MP uses ferromagnetic initialization by default.
This is correct; most calculations are initialized with a ferromagnetic ordering. This is how it was done historically, and is still how it is done for most materials.
However, for some materials we have started systematically searching for magnetic ground states by attempting multiple initial magnetic orderings: ferromagnetic, antiferromagnetic, ferrimagnetic, etc. This is detailed in two papers in 2019 and in 2020. Mostly, we have only done this searching for a set of around ~5000 transition metal oxides at present. This data was also made available via a data dump here for convenience, as detailed in the second paper.
If my understanding is incorrect and the information under âMagnetic Orderingâ indicates the initialization strategy before geometrical relaxation, what is the correct way to think when I have materials in my data collection where the magnetic ordering is stated as an AFM?
This is incorrect. The âMagnetic Orderingâ information shows the final magnetic ordering (after electronic minimization) for the lowest energy magnetic configuration in our database. Hence, if we have attempted an AFM ordering, and the AFM ordering is lower in energy than the FM ordering, this will state âAFMâ. Likewise, if a calculation was initialized as FM, but during electronic minimization became AFM, this will state âAFM.â
Hope this helps!
Matt
Hi @mkhorton
Thank you very much for your clarification! I have one follow up question/discuss point.
Given the default FM initialization strategy, I wonder whether a ML model trained on magnetic information from MP would be appropriate. One could argue that all possible initialization strategies werenât considered to find the âtrueâ lowest energy magnetic configuration (except for the ~5000 transition metal oxides mentioned in the 2020 paper). Therefore, the MP data is unsuitable for studies in magnetism.
On the other side, it also appears to make sense as long as the user of the model is aware of the default initialization strategy used by MP (as there are always bound to be some limitations in any computational data, whether this is due to some approximation or lack of strategies considered). And making use of the large data available in MP can still provide an edge in accelerating the discovery of magnetic materials, or they can be useful for testing a ML techniques. I have come across some ML papers that was published (before the 2020 paper you have mentioned above), which appears to use the MPâs magnetic ordering information as is.
Could I get your thoughts on this?
Many thanks,
Son
Therefore, the MP data is unsuitable for studies in magnetism. [âŚ] I have come across some ML papers that was published (before the 2020 paper you have mentioned above), which appears to use the MPâs magnetic ordering information as is.
It depends what youâre interested in; e.g. a distribution of magnitudes of atomic (site-projected) magnetic moments might still be relevant or useful to train on. If there have been studies that have tried to train classifiers of the magnetic ordering label (FM/FiM/AFM/etc), I think they would be in error, which is unfortunate but an easy mistake to make, unless they were limited to the materials that have been explicitly included in the magnetic ordering searches. Models trained on these labels may still contain some useful information, but probably not what was intended.
We are also continuing research on calculating magnetic orderings and intend to extend it far beyond the initial set of TMOs, but this is still in progress.
If this has caused confusion or problems for your own work, my apologies.
Best,
Matt