neighbors(method="gauss") returns a non-symmetric connectivity matrix on the sparse (pynndescent / transformer) path #4416

Please make sure these conditions are met

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of scanpy.
• (optional) I have confirmed this bug exists on the main branch of scanpy.

What happened?

The Gaussian kernel of Coifman & Lafon / Haghverdi et al. used by method="gauss" is symmetric, W[i, j] = W[j, i], and _connectivity.gauss explicitly symmetrises the kNN pattern ("compute the symmetric weight matrix"). sc.tl.diffmap / sc.tl.dpt build transitions_sym from these connectivities and assume a symmetric kernel.

Minimal code sample

import numpy as np, scanpy as sc
from anndata import AnnData

x = np.random.default_rng(0).normal(size=(9000, 10)).astype(np.float32)
for transformer in ("sklearn", None):        # None -> pynndescent for n_obs >= 8192
    a = AnnData(x.copy())
    sc.pp.neighbors(a, n_neighbors=15, method="gauss", use_rep="X",
                    transformer=transformer, random_state=0)
    w = a.obsp["connectivities"]
    print(transformer, "max|W - W.T| =", abs(w - w.T).max())

Error output

sklearn max|W - W.T| = 0.0
None max|W - W.T| = 0.65536994

Versions

scanpy 1.14.0
Python 3.13