Comparing MLIPs vs DFT for ionic conductivity in solid lithium electrolytes

### Comparing MLIPs vs DFT for ionic conductivity in solid lithium electrolytes

Alex Hernández-García (presenting)

.turquoise[[4th CEMDI–PAIMS Symposium 2026](https://cemdi.ca//symposium.php?id=evt_4th_symposium_2026) · Montréal · July 31st 2026]

.center[
<a href="https://mila.quebec/"><img src="../assets/images/slides/logos/mila-beige.png" alt="Mila" style="height: 4em"></a>
    
<a href="https://www.umontreal.ca/"><img src="../assets/images/slides/logos/udem-white.png" alt="UdeM" style="height: 4em"></a>
]

.center[
<a href="https://institut-courtois.umontreal.ca/"><img src="../assets/images/slides/logos/institut-courtois.png" alt="Institut Courtois" style="height: 2.5em"></a>
    
<a href="https://ivado.ca/"><img src="../assets/images/slides/logos/ivado.png" alt="IVADO" style="height: 2.5em"></a>
]

.smaller[.footer[
Slides: [alexhernandezgarcia.com/slides/{{ name }}](https://alexhernandezgarcia.com/slides/{{ name }})
]]

---

## Design of novel lithium electrolytes with high ionic conductivity for solid-state batteries

.center[
<figure>
	<img src="../assets/images/slides/ionic-conductivity/liquid_vs_solid_trans.png" alt="A cartoon depicting an illustration of a liquid battery and a solid-state battery side by side." style="width: 70%">
 <figcaption>.smaller[Adapted from: <a href="https://article.murata.com/en-sg/article/basic-lithium-ion-battery-4">Murata</a>]</figcaption>
</figure>
]

---

## Traditional screening

.center[
<figure>
	<img src="../assets/images/slides/ionic-conductivity/screening_he.png" alt="A typical funnel diagram for the application of discovering solid electrolytes with high ionic conductivity." style="width: 50%">
 <figcaption>.smaller[Source: <a href="https://www.nature.com/articles/s41597-020-0474-y">He et al. (2020)</a>]</figcaption>
</figure>
]

.references[He et al. [High-throughput screening platform for solid electrolytes combining hierarchical ion-transport prediction algorithms](https://www.nature.com/articles/s41597-020-0474-y). Scientific Data, 2020.]

---

## Screening guided by machine learning

.center[
<figure>
	<img src="../assets/images/slides/ionic-conductivity/screening_sendek.png" alt="A flowchart diagram representing a workflow for the application of discovering solid electrolytes with high ionic conductivity, using machine learning in the flow." style="width: 55%">
 <figcaption>.smaller[Source: <a href="https://pubs.acs.org/cmatex/article-abstract/31/2/342/1287786/Machine-Learning-Assisted-Discovery-of-Solid-Li">Sendek et al. (2018)</a>]</figcaption>
</figure>
]

.references[Sendek et al. [Machine learning-assisted discovery of solid Li-ion conducting materials](https://pubs.acs.org/cmatex/article-abstract/31/2/342/1287786/Machine-Learning-Assisted-Discovery-of-Solid-Li). Chemistry of Materials, 2018.]

---

## Our approach
### Iterative _de novo_ design with generative models and active learning

.left-column-33[
.highlight2[Iterative active learning]:
* mimics the .h1[traditional scientific discovery cycle],
* enables .h1[continuous improvement and refinement].
]

.left-column-33[
.highlight2[Generative machine learning] can:
* .highlight1[explore] unseen regions of the search space and
* propose .highlight1[novel candidates _de novo_].
]

---

## _Multi-fidelity_ active learning with generative modelling

.left-column-33[
 
.highlight1[Multi-fidelity active learning] can:
* leverage the availability of .highlight1[multiple oracles] with different .highlight1[costs and fidelity]
* efficiently use the right level of accuracy needed for each query
]

.references[Hernandez-Garcia, Saxena et al. [Multi-fidelity active learning with GFlowNets](https://arxiv.org/abs/2306.11715). TMLR, 2024]

---

## _Multi-fidelity_ active learning with generative modelling

.conclusion-float[Multi-fidelity active learning can leverage the diversity of methods available in science.]
]

.references[Hernandez-Garcia, Saxena et al. [Multi-fidelity active learning with GFlowNets](https://arxiv.org/abs/2306.11715). TMLR, 2024]

---

## Design of novel lithium electrolytes with high ionic conductivity for solid-state batteries

---

## Generative model
### Crystal-GFN

.right-column[.center[![:scale 100%](/assets/images/slides/scientific-discovery/gray/loop_4_highlight_genml.png)]]

.left-column[
.h1[Generative ML model]:
- The generated candidates should:
    - have the target property
    - have other desirable properties (lithium, available elements, etc.)
    - be realistic (physical constraints)
    - be diverse
- The available data sets are extremely small for ML standards.
]

---

## Crystal-GFlowNet

.conclusion-float[Crystal-GFN is a sequential generative model of crystal structures, to design candidates with desirable properties and constraints and inspired by theoretical crystallography.]

.references[Mila AI4Science et al. [Crystal-GFN: sampling crystals with desirable properties and constraints](https://arxiv.org/abs/2310.04925). AI4Mat, NeurIPS 2023 (spotlight)]

---

## Data
### OBELiX

.right-column-66[.center[![:scale 80%](/assets/images/slides/scientific-discovery/gray/loop_3_highlight_data.png)]]

.left-column-33[
.h1[Data]:
* Scarce and scattered data
* Especially of high quality
* Not ready for ML use
]

---

## OBELiX
### Open solid Battery Electrolytes with Li: an eXperimental dataset

.left-column-33[
We curated a data set of nearly 600 materials with experimentally measured ionic conductivity.
 
.center[![:scale 80%](/assets/images/slides/ionic-conductivity/obelix_venn.png)]
]

.full-width[
- Paper: [arxiv.org/abs/2502.14234](https://arxiv.org/abs/2502.14234)
- Code and data set: [github.com/NRC-Mila/OBELiX](https://github.com/NRC-Mila/OBELiX/tree/main)
]

---

## OBELiX
### Open solid Battery Electrolytes with Li: an eXperimental dataset

- Ionic conducitivities at room temperature
- Composition, space groups and lattice parameters for all 599 materials
- CIF files (structure) for 321 materials
- Strict train, validation, test splits to avoid data leakage

.references[
- Paper: [arxiv.org/abs/2502.14234](https://arxiv.org/abs/2502.14234)
- Code and data set: [github.com/NRC-Mila/OBELiX](https://github.com/NRC-Mila/OBELiX/tree/main)
]

---

## ML predictive model

.right-column[.center[![:scale 100%](/assets/images/slides/scientific-discovery/gray/loop_3_highlight_predml.png)]]

.left-column[
.h1[ML predictive model]:
- Random forest
- Multi-layer perceptron
- Graph neural networks
- ...
]

---

## ML predictive model

.right-column[.center[![:scale 100%](/assets/images/slides/scientific-discovery/gray/loop_3_highlight_predml.png)]]

.left-column[
.h1[ML predictive model]:
- Random forest
- Multi-layer perceptron
- Graph neural networks
- ...
 
.center[![:scale 100%](/assets/images/slides/ionic-conductivity/obelix_benchmark_fig.png)]
]

---

## Oracle

.right-column-66[.center[![:scale 80%](/assets/images/slides/scientific-discovery/gray/loop_4_highlight_oracle.png)]]

.left-column-33[
.h1[Oracle]: experimental validation
- Requires synthesising the material 
- Very high financial cost and multiple months per candidate.
]

---

## Oracle.h1[s]

.right-column-66[.center[![:scale 100%](/assets/images/slides/scientific-discovery/gray/loop_4_mf_highlight_oracles.png)]]

---

## Oracle.h1[s]

.right-column-66[.center[![:scale 100%](/assets/images/slides/scientific-discovery/gray/loop_4_mf_highlight_oracles.png)]]

.conclusion-float[What is the accuracy and the cost of DFT and MLIPs to estimate the ionic conductivity?]
]

---

## Ionic conductivity estimation
### A comparison of methods

.context[DFT has been widely used to estimate the ionic conductivity and recently MLIPs are widely available too.]

.left-column-33[.center[![:scale 100%](/assets/images/slides/ionic-conductivity/accuracy_time_methods_questionmark.png)]

Paper: [arxiv.org/abs/2603.28012](https://arxiv.org/abs/2603.28012)
]

.right-column-66[.center[![:scale 95%](/assets/images/slides/ionic-conductivity/comparative_study_paper.png)]]

---

## Methodology

- .h1[Goal]: Benchmark molecular dynamics (MD) approaches to estimate the ionic conductivity of lithium solid-state electrolytes.

--
    - A guiding objective was to also establish a framework that enables systematic comparisons across additional calculators and materials.
--
- .h1[MD methods]: 
    - DFT
    - MLIPs: NequIP, MACE and CHGNet
--
- .h1[Data]: 76 lithium solid-state electrolytes from [OBELiX data set](https://pubs.rsc.org/en/content/articlepdf/2026/dd/d5dd00441a), with experimentally measured ionic conductivity.

--
    - Structures without disorder
    - DFT was evaluated on a subset of 21 structures.

---