AQCat Adsorption Spin

​

Overview

​

What is AQCat Adsorption Spin?

• Specify a bulk material by composition string (e.g., "Co-Ni") — no Materials Project ID or crystal file required.
• Receive the lowest binding energy and per-placement convergence status within minutes, not hours.
• Screen catalyst configurations at a predictable, per-RU price.
• Integrate AQCat outputs directly into LLM-driven research workflows via Claude or other MCP clients.

​

How It Works

1. Bulk Selection: The system queries an internal database of crystal structures. If you specify a composition string, it scores candidates by element fraction and selects the highest-scoring, lowest-energy bulk automatically. Explicit Materials Project IDs (bulk_src_ids) and internal database indices (bulk_ids) are also supported.
2. Slab Generation: A surface slab is constructed from the selected bulk using the specified Miller indices (facet).
3. Adsorbate Placement: The adsorbate molecule is placed in up to 5 distinct configurations on the slab surface.
4. Geometry Relaxation: Each placement is passed through the AQCat25-EV2 MLP for iterative geometry relaxation. The MLP computes energies and forces at each step until convergence or the maximum step limit is reached.
5. Result Delivery: Results are delivered as a predictions array in the poll response — one entry per placement, each containing the binding energy (energy in eV), convergence status (is_relaxed), and per-atom force convergence (max_force).

​

Using AQCat

​

Setup

​

Improving Tool-Invocation Reliability

• Disable lazy loading in Claude Desktop — the single most impactful fix. See MCP → Improving tool-invocation reliability for the exact toggle.
• Name the tool explicitly in your prompt: “Use the AQCat tool to…” or “Call aqcat with…”.
• Include the surface facet and bulk composition in the prompt — e.g. “Cu (110)” or “Co-Ni (111)”. Ambiguous phrasing like “on copper” without a facet is more likely to be answered from prior training.
• Ask a scoped, computational question — “What is the binding energy of…” or “Find the lowest-energy configuration for…” triggers tool selection more reliably than “How does X adsorb on Y?”
• If your client still answers without calling the tool, explicitly ask it to. “Please call the aqcat tool with the parameters above” is a valid recovery prompt.

​

Example Prompts

AQCat evaluated 5 placements for N₂ on Co-Ni (111) (bulk reference `mp-1183837`); the lowest-energy configuration is placement 4 with a **binding energy of approximately −0.43 eV**. All returned placements relaxed to convergence within the step budget (`is_relaxed: true`).

AQCat evaluated 5 placements for CO on Cu (110) (bulk reference `mp-30`); all 5 relaxed to convergence within 100 steps and the lowest-energy configuration is placement 1 with a **binding energy of approximately −0.74 eV**.

AQCat H adsorption results across three Fe facets (bulk reference `mp-13`, 5 placements each, all relaxed to convergence):

Fe(110): **−0.50 eV** (strongest binding)
Fe(111): **−0.32 eV**
Fe(100): **−0.22 eV** (weakest of the three)

Fe(110) is the strongest H binder of the three and the most active candidate for hydrogen evolution — recommend Fe(110) for further validation.

AQCat CO adsorption results across four Cu-based compositions on the (111) facet (5 placements each, all relaxed to convergence):

Cu-Ni: **−1.28 eV** (strongest binding)
Cu-Pd: **−1.12 eV**
Cu: **−0.55 eV**
Cu-Zn: **−0.38 eV** (weakest of the four)

All four use the same CO gas-phase reference, so the ranking is valid. Cu-Ni and Cu-Pd bind CO substantially more strongly than pure Cu — likely too strongly for a CO oxidation catalyst, where the volcano peak favours intermediate binding.

AQCat N adsorption results (5 placements each, all relaxed to convergence):

Fe (110) (bulk reference `mp-13`): **−0.86 eV**, `is_spin_off: false` — spin polarization enabled
Pt (111) (bulk reference `mp-126`): **−0.31 eV**, `is_spin_off: true` — spin polarization disabled

Fe is in the spin-polarized element set (Ce, Co, Cr, Cu, Fe, Mn, Mo, Ni, Os, Ru, V, W), so its training data carries magnetic structure; Pt is not. Both energies use the same N₂ gas-phase reference, so comparing the two surfaces is valid — this is a same-adsorbate comparison, which is the only kind AQCat energies support.

​

Request & Response Schemas

​

Request Schema

{
  "client_context_id": "user_internal_project_01",
  "input_type": "dense_adsorption_workflow",
  "system_parameters": {
    "bulk_composition_contains": "Co-Ni",
    "max_bulks": 1,
    "adsorbates": ["*N2"],
    "facets": [[1, 1, 1]]
  },
  "workflow_parameters": {
    "max_steps": 100,
    "placements": 5
  }
}

​

Response Schema

{
  "job_id": "955ac259-ecab-4b9f-9f1a-abb347c4b863",
  "tool_name": "query_aqcat",
  "status": "running",
  "is_terminal": false,
  "progress_message": "Running... (45s elapsed)",
  "elapsed_seconds": 45,
  "predictions": []
}

{
  "job_id": "955ac259-ecab-4b9f-9f1a-abb347c4b863",
  "tool_name": "aqcat",
  "detail": false,
  "predictions": [
    {
      "combo_key": "CO_mp-30_110_125_True",
      "placement_idx": 1,
      "energy": -0.7388379573822021,
      "max_force": 0.03282981365919113,
      "is_relaxed": true,
      "is_spin_off": false,
      "is_low_fi": false
    }
  ]
}

​

Supported Input Formats

​

Supported Inputs Reference

​

Interpreting Results

​

Energy Sign Convention & Reference States

​

The is_spin_off flag

​

Scientific Methodology

​

The AQCat25 Dataset

• Spin polarization for Ce, Co, Cr, Cu, Fe, Mn, Mo, Ni, Os, Ru, V, and W. AQCat25 is the only large-scale public dataset carrying spin-polarized data across a diverse heterogeneous catalyst space.
• A 500 eV plane-wave cutoff, up from 350 eV in OC20, with Gaussian smearing at 0.1 eV. Non-metal catalysts need the higher cutoff for accurate total energies.
• Six elements OC20 never covered: lithium, barium, lanthanum, cerium, magnesium, and fluorine, along with 20 reaction transition states.

​

Model & Dataset Availability

​

Pricing

​

Limits & Quotas

​

Security & IP Protection

• Telemetry: Per-call usage metadata (counts, identifiers, wall-clock time) is logged for billing reconciliation and capacity planning.
• Privacy: See the SandboxAQ Privacy Policy for complete details on data handling and retention.
• Organizational isolation: Usage, billing, and data are isolated per organization. Cross-organization access is not possible.

​

Figures & References

• Allam, O., Wander, B., Kim, S., et al. “AQCat25: Unlocking spin-aware, high-fidelity machine learning potentials for heterogeneous catalysis.” arXiv:2510.22938v2
• Figure 7 of arXiv:2510.22938v2 — parity plots of DFT vs. ML minimum adsorption energies for the pretrained EV2 baseline and AQCat25-EV2, annotated with MAE and success rate.
• Figure 1 of arXiv:2510.22938v2 — AQCat25 dataset composition and the multi-fidelity training approach.
• SandboxAQ blog, “Introducing our AQCat25-EV2 Models for AI-Accelerated Catalysis Discovery,” October 27, 2025.