# ICT-CSP

An open-source Python framework that integrates machine learning interatomic
potentials (MLIPs) with a tailored batched optimization strategy, enabling rapid,
unbiased structure prediction across the full density range.

## Installation (Hygon DCU)

### 1. Clone the repository

```sh
git clone -b public git@gitee.com:nivren/ict-csp.git
cd ict-csp
```

### 2. Configure local torch & triton wheels

Edit `pyproject.toml` — replace the paths under `[tool.uv.sources]` for `torch` and
`triton` with the local `.whl` file paths for your DCU platform.

### 3. Fetch third-party submodules

```sh
git submodule update --init \
  batchASE \
  3rdparty/mace \
  3rdparty/FastEq \
  3rdparty/cuequivariance_torch
```

### 4. Set up build dependencies

```sh
source /opt/dtk/env.sh

sudo apt install python3.10-dev python3-dev
sudo apt install libgflags-dev libgoogle-glog-dev
sudo apt install build-essential libstdc++-12-dev
sudo ln -s /usr/lib/x86_64-linux-gnu/librt.so.1 /usr/lib/x86_64-linux-gnu/librt.so
```

### 5. Create the virtual environment

> **Prerequisite:** This project uses [uv](https://docs.astral.sh/uv/) for Python
> environment management. Install it first if you haven't:
> ```sh
> curl -LsSf https://astral.sh/uv/install.sh | sh
> ```

```sh
uv sync
```

### 6. Verify the installation

```sh
uv run python -c "import torch, ase, mace, fasteq; print('ok')"
```

### 7. Prepare MACE model weights

Ensure `~/.cache/mace/` contains the required weight file (e.g.
`MACE-OFF23_small.model`). Download from:

- <https://github.com/ACEsuit/mace-off/tree/main>
- <https://gitee.com/pfsuo/mace-off/raw/main/mace_off23/MACE-OFF23_small.model>

### 8. Set up a test run directory

```sh
mkdir run && cp example/* run/
```

### 9. Test crystal structure generation

```sh
source .venv/bin activate
./run/generation.sh
```

### 10. Test crystal structure relaxation

```sh
source .venv/bin/activate
./run/relaxation.sh
```

## Results & Configuration

- **Results** are written to the `csp_results/` directory.
- **Configuration files** for generation and relaxation are located in the `run/` directory.

## Tested Environment

Hygon DCU · BW200 · DTK 25.04.1