cuRDF 0.5.0__tar.gz → 0.5.2__tar.gz

{curdf-0.5.0 → curdf-0.5.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cuRDF
-Version: 0.5.0
+Version: 0.5.2
 Summary: GPU-accelerated radial distribution functions with NVIDIA ALCHEMI Toolkit-Ops neighbor lists and PyTorch.
 Author: Joseph Hart
 License-Expression: MIT
@@ -45,7 +45,9 @@ Requires-Dist: sphinx-rtd-theme; extra == "docs"
 Requires-Dist: myst-parser; extra == "docs"
 Dynamic: license-file
 
-# cuRDF
+![cuRDF logo](logo.svg)
+
+---
 
 [![DOI](https://zenodo.org/badge/1085332119.svg)](https://doi.org/10.5281/zenodo.1085332119) [![PyPI](https://img.shields.io/pypi/v/cuRDF.svg)](https://pypi.org/project/cuRDF/)
 
@@ -56,7 +58,8 @@ CUDA-accelerated radial distribution functions using [NVIDIA ALCHEMI Toolkit-Ops
 
 ![cuRDF benchmark](benchmarks/results/results.png)
 
-cuRDF is benchmarked against RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.
+cuRDF is benchmarked against other RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.
+Benchmarks use random positions in cubic cells sized to maintain a fixed number density (orthorhombic boxes, periodic in all directions).
 
 ## Install
 Latest release:
@@ -114,9 +117,27 @@ bins, gr = rdf(
 
 If the topology lacks atom names (only numeric types), supply a mapping:
 ```python
-bins, gr = rdf(u, species_a="C", species_b="O", atom_types_map={1: "C", 2: "H"})
+bins, gr = rdf(
+  u, 
+  species_a="C", 
+  species_b="O", 
+  atom_types_map={1: "C", 2: "H"}
+)
 ```
 
+## Validation
+
+RDFs for liquid water (64 atoms, 1 ns) computed using a trajectory from [Lim et al. (2026)](https://arxiv.org/abs/2601.20134) match reference curves for all pairs:
+
+<table>
+  <tr>
+    <td><img src="tests/cuRDF_results/rdf_OO_comparison.svg" alt="O–O RDF" width="250"></td>
+    <td><img src="tests/cuRDF_results/rdf_HH_comparison.svg" alt="H–H RDF" width="250"></td>
+    <td><img src="tests/cuRDF_results/rdf_OH_comparison.svg" alt="O–H RDF" width="250"></td>
+  </tr>
+</table>
+
+
 ## Citation
 If you use cuRDF in your work, please cite:
 ```
@@ -126,7 +147,7 @@ If you use cuRDF in your work, please cite:
   month     = dec,
   year      = 2025,
   publisher = {Zenodo},
-  version   = {0.1.0},
+  version   = {0.5.0},
   doi       = {10.5281/zenodo.1085332119},
   url       = {https://doi.org/10.5281/zenodo.1085332119}
 }

{curdf-0.5.0 → curdf-0.5.2}/README.md

@@ -1,4 +1,6 @@
-# cuRDF
+![cuRDF logo](logo.svg)
+
+---
 
 [![DOI](https://zenodo.org/badge/1085332119.svg)](https://doi.org/10.5281/zenodo.1085332119) [![PyPI](https://img.shields.io/pypi/v/cuRDF.svg)](https://pypi.org/project/cuRDF/)
 
@@ -9,7 +11,8 @@ CUDA-accelerated radial distribution functions using [NVIDIA ALCHEMI Toolkit-Ops
 
 ![cuRDF benchmark](benchmarks/results/results.png)
 
-cuRDF is benchmarked against RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.
+cuRDF is benchmarked against other RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.
+Benchmarks use random positions in cubic cells sized to maintain a fixed number density (orthorhombic boxes, periodic in all directions).
 
 ## Install
 Latest release:
@@ -67,9 +70,27 @@ bins, gr = rdf(
 
 If the topology lacks atom names (only numeric types), supply a mapping:
 ```python
-bins, gr = rdf(u, species_a="C", species_b="O", atom_types_map={1: "C", 2: "H"})
+bins, gr = rdf(
+  u, 
+  species_a="C", 
+  species_b="O", 
+  atom_types_map={1: "C", 2: "H"}
+)
 ```
 
+## Validation
+
+RDFs for liquid water (64 atoms, 1 ns) computed using a trajectory from [Lim et al. (2026)](https://arxiv.org/abs/2601.20134) match reference curves for all pairs:
+
+<table>
+  <tr>
+    <td><img src="tests/cuRDF_results/rdf_OO_comparison.svg" alt="O–O RDF" width="250"></td>
+    <td><img src="tests/cuRDF_results/rdf_HH_comparison.svg" alt="H–H RDF" width="250"></td>
+    <td><img src="tests/cuRDF_results/rdf_OH_comparison.svg" alt="O–H RDF" width="250"></td>
+  </tr>
+</table>
+
+
 ## Citation
 If you use cuRDF in your work, please cite:
 ```
@@ -79,7 +100,7 @@ If you use cuRDF in your work, please cite:
   month     = dec,
   year      = 2025,
   publisher = {Zenodo},
-  version   = {0.1.0},
+  version   = {0.5.0},
   doi       = {10.5281/zenodo.1085332119},
   url       = {https://doi.org/10.5281/zenodo.1085332119}
 }

{curdf-0.5.0 → curdf-0.5.2}/cuRDF.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cuRDF
-Version: 0.5.0
+Version: 0.5.2
 Summary: GPU-accelerated radial distribution functions with NVIDIA ALCHEMI Toolkit-Ops neighbor lists and PyTorch.
 Author: Joseph Hart
 License-Expression: MIT
@@ -45,7 +45,9 @@ Requires-Dist: sphinx-rtd-theme; extra == "docs"
 Requires-Dist: myst-parser; extra == "docs"
 Dynamic: license-file
 
-# cuRDF
+![cuRDF logo](logo.svg)
+
+---
 
 [![DOI](https://zenodo.org/badge/1085332119.svg)](https://doi.org/10.5281/zenodo.1085332119) [![PyPI](https://img.shields.io/pypi/v/cuRDF.svg)](https://pypi.org/project/cuRDF/)
 
@@ -56,7 +58,8 @@ CUDA-accelerated radial distribution functions using [NVIDIA ALCHEMI Toolkit-Ops
 
 ![cuRDF benchmark](benchmarks/results/results.png)
 
-cuRDF is benchmarked against RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.
+cuRDF is benchmarked against other RDF (MDAnalysis) and neighbour list implementations on CPU (AMD Ryzen 9 9950X, 32 threads) and GPU (NVIDIA RTX 5090) for systems of varying sizes at a density of 0.05 atoms/ų over 1000 frames.
+Benchmarks use random positions in cubic cells sized to maintain a fixed number density (orthorhombic boxes, periodic in all directions).
 
 ## Install
 Latest release:
@@ -114,9 +117,27 @@ bins, gr = rdf(
 
 If the topology lacks atom names (only numeric types), supply a mapping:
 ```python
-bins, gr = rdf(u, species_a="C", species_b="O", atom_types_map={1: "C", 2: "H"})
+bins, gr = rdf(
+  u, 
+  species_a="C", 
+  species_b="O", 
+  atom_types_map={1: "C", 2: "H"}
+)
 ```
 
+## Validation
+
+RDFs for liquid water (64 atoms, 1 ns) computed using a trajectory from [Lim et al. (2026)](https://arxiv.org/abs/2601.20134) match reference curves for all pairs:
+
+<table>
+  <tr>
+    <td><img src="tests/cuRDF_results/rdf_OO_comparison.svg" alt="O–O RDF" width="250"></td>
+    <td><img src="tests/cuRDF_results/rdf_HH_comparison.svg" alt="H–H RDF" width="250"></td>
+    <td><img src="tests/cuRDF_results/rdf_OH_comparison.svg" alt="O–H RDF" width="250"></td>
+  </tr>
+</table>
+
+
 ## Citation
 If you use cuRDF in your work, please cite:
 ```
@@ -126,7 +147,7 @@ If you use cuRDF in your work, please cite:
   month     = dec,
   year      = 2025,
   publisher = {Zenodo},
-  version   = {0.1.0},
+  version   = {0.5.0},
   doi       = {10.5281/zenodo.1085332119},
   url       = {https://doi.org/10.5281/zenodo.1085332119}
 }

{curdf-0.5.0 → curdf-0.5.2}/cuRDF.egg-info/SOURCES.txt

@@ -14,4 +14,5 @@ curdf/plotting.py
 curdf/rdf.py
 tests/test_rdf_core.py
 tests/test_rdf_cuda.py
-tests/test_rdf_sources.py
+tests/test_rdf_sources.py
+tests/test_water_rdfs.py

{curdf-0.5.0 → curdf-0.5.2}/curdf/adapters.py

@@ -50,9 +50,14 @@ def rdf_from_mdanalysis(
     selection_b: str | None = None,
     atom_types_map: dict | None = None,
     index=None,
+    start: int | None = None,
+    stop: int | None = None,
+    step: int | None = None,
     r_min: float = 1.0,
     r_max: float = 6.0,
     nbins: int = 100,
+    r_min_floor: float = 1e-6,
+    pbc: tuple[bool, bool, bool] = (True, True, True),
     device="cuda",
     torch_dtype=None,
     half_fill: bool = True,
@@ -77,12 +82,22 @@ def rdf_from_mdanalysis(
         Optional mapping for numeric atom types to element names (e.g., ``{1: "C", 2: "H"}``).
     index
         Optional trajectory index/selector.
+    start
+        Optional starting frame (inclusive) when iterating the trajectory.
+    stop
+        Optional stopping frame (exclusive) when iterating the trajectory.
+    step
+        Optional frame stride when iterating the trajectory.
     r_min
         Minimum distance included in the histogram.
     r_max
         Maximum distance included in the histogram.
     nbins
         Number of histogram bins.
+    r_min_floor
+        Small lower bound to exclude pathological self/near-zero distances when ``r_min`` is zero.
+    pbc
+        Tuple of periodic boundary flags for (a, b, c) directions.
     device
         Torch device string or object used for computation.
     torch_dtype
@@ -149,21 +164,26 @@ def rdf_from_mdanalysis(
         raise ValueError(f"No atoms found for species_a='{species_a}' (check atom names or --atom-types mapping)")
     if len(ag_b) == 0:
         raise ValueError(f"No atoms found for species_b='{species_b or species_a}' (check atom names or --atom-types mapping)")
-    if wrap_positions and mda_wrap is not None:
+    if wrap_positions and mda_wrap is not None and all(pbc):
         ag_wrap = ag_a if selection_b is None else (ag_a | ag_b)
         universe.trajectory.add_transformations(mda_wrap(ag_wrap, compound="atoms"))
 
     same_species = len(ag_a) == len(ag_b) and ag_a is ag_b
 
     def frames():
-        traj = universe.trajectory[index] if index is not None else universe.trajectory
+        if index is not None:
+            traj = universe.trajectory[index]
+        elif start is not None or stop is not None or step is not None:
+            traj = universe.trajectory[start:stop:step]
+        else:
+            traj = universe.trajectory
         for ts in tqdm(traj, desc="Frames (MDAnalysis)", unit="frame"):
             cell = _mdanalysis_cell_matrix(ts.dimensions)
             if same_species:
                 yield {
                     "positions": ag_a.positions.astype(np.float32, copy=False),
                     "cell": cell,
-                    "pbc": (True, True, True),
+                    "pbc": tuple(pbc),
                 }
             else:
                 pos_a = ag_a.positions.astype(np.float32, copy=False)
@@ -176,7 +196,7 @@ def rdf_from_mdanalysis(
                 yield {
                     "positions": pos,
                     "cell": cell,
-                    "pbc": (True, True, True),
+                    "pbc": tuple(pbc),
                     "group_a_mask": group_a_mask,
                     "group_b_mask": group_b_mask,
                 }
@@ -189,6 +209,7 @@ def rdf_from_mdanalysis(
         r_min=r_min,
         r_max=r_max,
         nbins=nbins,
+        r_min_floor=r_min_floor,
         device=device,
         torch_dtype=torch_dtype,
         half_fill=half_fill,
@@ -234,6 +255,8 @@ def rdf_from_ase(
     r_min: float = 1.0,
     r_max: float = 6.0,
     nbins: int = 100,
+    r_min_floor: float = 1e-6,
+    pbc: tuple[bool, bool, bool] = (True, True, True),
     device="cuda",
     torch_dtype=None,
     half_fill: bool = True,
@@ -266,6 +289,10 @@ def rdf_from_ase(
         Maximum distance included in the histogram.
     nbins
         Number of histogram bins.
+    r_min_floor
+        Small lower bound to exclude pathological self/near-zero distances when ``r_min`` is zero.
+    pbc
+        Tuple of periodic boundary flags for (a, b, c) directions.
     device
         Torch device string or object used for computation.
     torch_dtype
@@ -308,7 +335,7 @@ def rdf_from_ase(
 
             if atom_types_map:
                 # map numeric types to element names if provided
-                types = frame.get_array("numbers", int, None)
+                types = frame.get_array("numbers", False)
                 name_list = []
                 for t in types:
                     name = atom_types_map.get(t) or atom_types_map.get(str(t))
@@ -340,24 +367,38 @@ def rdf_from_ase(
                     "Check element symbols or index selection."
                 )
 
-            pos_all = frame.get_positions(wrap=wrap_positions)
-            pos_a = pos_all[idx_a]
-            pos_b = pos_all[idx_b]
-            pos = np.concatenate([pos_a, pos_b], axis=0)
+            if isinstance(pbc, Iterable):
+                pbc_tuple = tuple(bool(x) for x in pbc)
+            else:
+                pbc_tuple = (bool(pbc), bool(pbc), bool(pbc))
+            wrap_flag = wrap_positions and all(pbc_tuple)
+            pos_all = frame.get_positions(wrap=wrap_flag)
             cell = np.array(frame.get_cell().array, dtype=np.float32)
-            pbc = tuple(bool(x) for x in frame.get_pbc())
-            group_a_mask = np.zeros(len(pos), dtype=bool)
-            group_b_mask = np.zeros(len(pos), dtype=bool)
-            group_a_mask[: len(pos_a)] = True
-            group_b_mask[len(pos_a) :] = True
-
-            yield {
-                "positions": pos.astype(np.float32, copy=False),
-                "cell": cell,
-                "pbc": pbc,
-                "group_a_mask": group_a_mask,
-                "group_b_mask": group_b_mask,
-            }
+            
+            if same_species:
+                # Same-species: keep one group so half_fill + pair_factor=2 normalisation stays correct
+                yield {
+                    "positions": pos_all[idx_a].astype(np.float32, copy=False),
+                    "cell": cell,
+                    "pbc": pbc_tuple,
+                }
+            else:
+                # Cross-species: concatenate and mask groups for ordered pairs
+                pos_a = pos_all[idx_a]
+                pos_b = pos_all[idx_b]
+                pos = np.concatenate([pos_a, pos_b], axis=0)
+                group_a_mask = np.zeros(len(pos), dtype=bool)
+                group_b_mask = np.zeros(len(pos), dtype=bool)
+                group_a_mask[: len(pos_a)] = True
+                group_b_mask[len(pos_a) :] = True
+
+                yield {
+                    "positions": pos.astype(np.float32, copy=False),
+                    "cell": cell,
+                    "pbc": pbc_tuple,
+                    "group_a_mask": group_a_mask,
+                    "group_b_mask": group_b_mask,
+                }
 
     same_species = (
         (species_b is None or species_b == species_a)
@@ -371,6 +412,7 @@ def rdf_from_ase(
         r_min=r_min,
         r_max=r_max,
         nbins=nbins,
+        r_min_floor=r_min_floor,
         device=device,
         torch_dtype=torch_dtype,
         half_fill=half_fill,

{curdf-0.5.0 → curdf-0.5.2}/curdf/rdf.py

@@ -10,6 +10,8 @@ from torch import Tensor
 from .cell import cell_tensor, cell_volume, pbc_tensor
 from .neighbor import build_neighbor_list
 
+DEFAULT_R_MIN_FLOOR = 1e-6
+
 
 def _update_counts(
     counts: Tensor,
@@ -22,6 +24,7 @@ def _update_counts(
     half_fill: bool,
     max_neighbors: int | None,
     method: str,
+    r_min_floor: float = DEFAULT_R_MIN_FLOOR,
     group_a_mask: Tensor | None = None,
     group_b_mask: Tensor | None = None,
 ) -> float:
@@ -44,6 +47,8 @@ def _update_counts(
         Minimum distance included in the histogram.
     r_max
         Maximum distance included in the histogram.
+    r_min_floor
+        Small lower bound to exclude pathological self/near-zero distances when ``r_min`` is zero.
     half_fill
         Whether to build unique pairs (same-species mode).
     max_neighbors
@@ -79,7 +84,10 @@ def _update_counts(
     dr_vec = (positions[tgt] + shift_cart) - positions[src]
     dist = torch.linalg.norm(dr_vec, dim=1)
 
-    valid = (dist >= r_min) & (dist < r_max)
+    r_min_eff = max(r_min, r_min_floor)
+    valid = (dist >= r_min_eff) & (dist < r_max)
+    # Drop any self-pairs that may appear in the neighbor list (important when r_min == 0).
+    valid = valid & (src != tgt)
     if group_a_mask is not None and group_b_mask is not None:
         src_mask = group_a_mask[src]
         tgt_mask = group_b_mask[tgt]
@@ -91,13 +99,19 @@ def _update_counts(
     counts.scatter_add_(0, bin_idx, torch.ones_like(bin_idx, dtype=torch.int64))
 
     volume = cell_volume(cell)
+
     if group_a_mask is not None and group_b_mask is not None:
         n_a = group_a_mask.sum().item()
         n_b = group_b_mask.sum().item()
-        norm_factor = n_a * (n_b / volume)
+        # Check if same species
+        if torch.equal(group_a_mask, group_b_mask):
+            norm_factor = n_a * ((n_b - 1) / volume)  # Exclude self-pairs
+        else:
+            norm_factor = n_a * (n_b / volume)  # Different species
     else:
         n_atoms = positions.shape[0]
-        norm_factor = n_atoms * (n_atoms / volume)  # n_atoms * rho
+        norm_factor = n_atoms * ((n_atoms - 1) / volume)  # Exclude self-pairs
+
     return norm_factor
 
 
@@ -149,6 +163,7 @@ def compute_rdf(
     r_min: float = 1.0,
     r_max: float = 6.0,
     nbins: int = 100,
+    r_min_floor: float = DEFAULT_R_MIN_FLOOR,
     device: str | torch.device = "cuda",
     torch_dtype: torch.dtype = torch.float32,
     half_fill: bool = True,
@@ -174,6 +189,8 @@ def compute_rdf(
         Maximum distance included in the histogram.
     nbins
         Number of histogram bins.
+    r_min_floor
+        Small lower bound to exclude pathological self/near-zero distances when ``r_min`` is zero.
     device
         Torch device string or object used for computation.
     torch_dtype
@@ -202,7 +219,8 @@ def compute_rdf(
     cell_t = cell_tensor(cell, device=device, dtype=torch_dtype)
     pbc_t = pbc_tensor(pbc, device=device)
 
-    edges = torch.linspace(r_min, r_max, nbins + 1, device=device, dtype=torch_dtype)
+    r_min_eff = max(r_min, r_min_floor)
+    edges = torch.linspace(r_min_eff, r_max, nbins + 1, device=device, dtype=torch_dtype)
     counts = torch.zeros(nbins, device=device, dtype=torch.int64)
 
     group_a_mask = group_b_mask = None
@@ -215,26 +233,29 @@ def compute_rdf(
     elif group_a_mask is not None:
         group_b_mask = group_a_mask
 
+    cross_mode = group_a_mask is not None and group_b_mask is not None and not torch.equal(
+        group_a_mask, group_b_mask
+    )
+    half_fill_eff = False if cross_mode else half_fill
+
     total_norm = _update_counts(
         counts,
         pos_t,
         cell=cell_t,
         pbc=pbc_t,
         edges=edges,
-        r_min=r_min,
+        r_min=r_min_eff,
         r_max=r_max,
-        half_fill=half_fill,
+        half_fill=half_fill_eff,
         max_neighbors=max_neighbors,
         method=method,
+        r_min_floor=r_min_floor,
         group_a_mask=group_a_mask,
         group_b_mask=group_b_mask,
     )
 
-    cross_mode = group_a_mask is not None and group_b_mask is not None and not torch.equal(
-        group_a_mask, group_b_mask
-    )
     centers, g_r = _finalize_gr(
-        counts, edges, total_norm, half_fill=half_fill, cross_mode=cross_mode
+        counts, edges, total_norm, half_fill=half_fill_eff, cross_mode=cross_mode
     )
     return centers.cpu().numpy(), g_r.cpu().numpy()
 
@@ -245,6 +266,7 @@ def accumulate_rdf(
     r_min: float,
     r_max: float,
     nbins: int,
+    r_min_floor: float,
     device: str | torch.device,
     torch_dtype: torch.dtype,
     half_fill: bool,
@@ -264,6 +286,8 @@ def accumulate_rdf(
         Maximum distance included in the histogram.
     nbins
         Number of histogram bins.
+    r_min_floor
+        Small lower bound to exclude pathological self/near-zero distances when ``r_min`` is zero.
     device
         Torch device string or object used for computation.
     torch_dtype
@@ -281,7 +305,8 @@ def accumulate_rdf(
         Bin centers and g(r) arrays on CPU.
     """
     device = torch.device(device)
-    edges = torch.linspace(r_min, r_max, nbins + 1, device=device, dtype=torch_dtype)
+    r_min_eff = max(r_min, r_min_floor)
+    edges = torch.linspace(r_min_eff, r_max, nbins + 1, device=device, dtype=torch_dtype)
     counts = torch.zeros(nbins, device=device, dtype=torch.int64)
     total_norm = 0.0
 
@@ -303,6 +328,9 @@ def accumulate_rdf(
             group_a_mask, group_b_mask
         ):
             cross_flag = True
+            half_fill_frame = False
+        else:
+            half_fill_frame = half_fill
 
         norm = _update_counts(
             counts,
@@ -310,21 +338,23 @@ def accumulate_rdf(
             cell=cell_t,
             pbc=pbc_t,
             edges=edges,
-            r_min=r_min,
+            r_min=r_min_eff,
             r_max=r_max,
-            half_fill=half_fill,
+            half_fill=half_fill_frame,
             max_neighbors=max_neighbors,
             method=method,
+            r_min_floor=r_min_floor,
             group_a_mask=group_a_mask,
             group_b_mask=group_b_mask,
         )
         total_norm += norm
 
+    half_fill_eff = False if cross_flag else half_fill
     centers, g_r = _finalize_gr(
         counts,
         edges,
         total_norm,
-        half_fill=half_fill,
+        half_fill=half_fill_eff,
         cross_mode=cross_flag,
     )
     return centers.cpu().numpy(), g_r.cpu().numpy()
@@ -336,6 +366,7 @@ def rdf(
     species_b: str | None = None,
     index=None,
     atom_types_map: dict | None = None,
+    pbc: tuple[bool, bool, bool] | None = None,
     method: str = "cell_list",
     outdir=None,
     output: str | None = None,
@@ -356,6 +387,8 @@ def rdf(
         Optional index/selector forwarded to source-specific readers.
     atom_types_map
         Optional mapping for numeric atom types to element names.
+    pbc
+        Optional periodic boundary flags; defaults to (True, True, True) when omitted.
     method
         Neighbor-list method name (e.g., ``"cell_list"`` or ``"naive"``).
     outdir
@@ -380,6 +413,7 @@ def rdf(
             species_b=species_b,
             index=index,
             atom_types_map=atom_types_map,
+            pbc=pbc if pbc is not None else (True, True, True),
             method=method,
             **kwargs,
         )
@@ -392,6 +426,7 @@ def rdf(
             species_b=species_b,
             index=index,
             atom_types_map=atom_types_map,
+            pbc=pbc if pbc is not None else (True, True, True),
             method=method,
             **kwargs,
         )
@@ -405,6 +440,7 @@ def rdf(
             species_b=species_b,
             index=index,
             atom_types_map=atom_types_map,
+            pbc=pbc if pbc is not None else (True, True, True),
             method=method,
             **kwargs,
         )

{curdf-0.5.0 → curdf-0.5.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "cuRDF"
-version = "0.5.0"
+version = "0.5.2"
 description = "GPU-accelerated radial distribution functions with NVIDIA ALCHEMI Toolkit-Ops neighbor lists and PyTorch."
 authors = [{ name = "Joseph Hart" }]
 readme = { file = "README.md", "content-type" = "text/markdown" }

curdf-0.5.2/tests/test_water_rdfs.py

@@ -0,0 +1,55 @@
+import numpy as np
+import pytest
+
+from curdf.rdf import rdf
+from ase.io import read
+import matplotlib.pyplot as plt
+from pathlib import Path
+
+DATAPATH = Path("darren_data/rdf-mace-nvt")
+RMIN = 0.0
+RMAX = 6.0
+N_BINS = 200
+
+def test_compute_rdf_cuda_runs():
+    torch_mod = pytest.importorskip("torch")
+    if not torch_mod.cuda.is_available():
+        pytest.skip("CUDA not available")
+    
+    water_traj_list = read(DATAPATH / f'nvt.xyz', index=':')
+    
+    for species_a, species_b in [("O", "O"), ("H", "H"), ("O", "H")]:
+        data = np.loadtxt(DATAPATH / f"rdf_{species_a}{species_b}_mace.dat")
+        r_ref, gr_ref = data[:,0], data[:,1]
+        
+        try:
+            bins, gr = rdf(
+                water_traj_list,
+                species_a=species_a,
+                species_b=species_b,
+                r_min=RMIN,
+                r_max=RMAX,
+                nbins=N_BINS,
+                output=f"cuRDF_results/rdf_{species_a}{species_b}.csv"
+            )
+            plt.plot(bins, gr, label=f"cuRDF - {species_a}-{species_b}")
+            plt.plot(r_ref, gr_ref, label=f"Reference - {species_a}-{species_b}")
+            plt.xlabel("r (Å)")
+            plt.ylabel("g(r)")
+            plt.title(f"RDF {species_a}-{species_b}")
+            plt.ylim(0,)
+            plt.xlim(0, RMAX)
+            plt.legend()
+            plt.tight_layout()
+            plt.savefig(f"cuRDF_results/rdf_{species_a}{species_b}_comparison.png", dpi=300)
+            plt.savefig(f"cuRDF_results/rdf_{species_a}{species_b}_comparison.pdf")
+            plt.savefig(f"cuRDF_results/rdf_{species_a}{species_b}_comparison.svg")
+            plt.close()
+            # Numerical check: bins should align and g(r) should match reference within 1% relative.
+            assert bins.shape == r_ref.shape
+            assert gr.shape == gr_ref.shape
+            np.testing.assert_allclose(bins, r_ref, rtol=1e-6, atol=1e-6)
+            np.testing.assert_allclose(gr, gr_ref, rtol=1e-2, atol=1e-3)
+        except RuntimeError as err:
+            pytest.skip(f"CUDA neighbor list unavailable: {err}")
+