potential function fixes

ferminet/hamiltonian.py

@@ -144,16 +144,15 @@ def grad_phase_closure(x):
   return _lapl_over_f
 
 
-def potential_electron_electron(pos: Array) -> jnp.ndarray:
+def potential_electron_electron(r_ee: Array) -> jnp.ndarray:
   """Returns the electron-electron potential.
 
   Args:
-    pos: Shape (neletrons, ndim). Electron positions.
+    r_ee: Shape (neletrons, nelectrons, :). r_ee[i,j,0] gives the distance
+      between electrons i and j. Other elements in the final axes are not
+      required.
   """
-  nelectrons = pos.shape[0]
-  r_ee = jnp.sqrt(jnp.sum((pos[None, ...] - pos[:, None] + 
-      jnp.eye(nelectrons)[..., None])**2, axis=-1))
-  r_ee = r_ee[jnp.triu_indices_from(r_ee, 1)]
+  r_ee = r_ee[jnp.triu_indices_from(r_ee[..., 0], 1)]
   return (1./r_ee).sum()
 
 
@@ -236,9 +235,8 @@ def _e_l(
       data: MCMC configuration.
     """
     del key  # unused
-    _, _, r_ae, _ = networks.construct_input_features(data.positions, data.atoms)
-    pos = data.positions.reshape(r_ae.shape[0],-1)
-    potential = potential_energy(r_ae, pos, data.atoms, charges)
+    _, _, r_ae, r_ee = networks.construct_input_features(data.positions, data.atoms)
+    potential = potential_energy(r_ae, r_ee, data.atoms, charges)
     kinetic = ke(params, data)
     return potential + kinetic
 

ferminet/networks.py

@@ -471,9 +471,8 @@ def construct_input_features(
   r_ae = jnp.linalg.norm(ae, axis=2, keepdims=True)
   # Avoid computing the norm of zero, as is has undefined grad
   n = ee.shape[0]
-  # r_ee = (
-  #     jnp.linalg.norm(ee + jnp.eye(n)[..., None], axis=-1) * (1.0 - jnp.eye(n)))
-  r_ee = jnp.sqrt(((ee + jnp.eye(n)[..., None])**2).sum(2)) * (1.0 - jnp.eye(n))
+  r_ee = (
+      jnp.linalg.norm(ee + jnp.eye(n)[..., None], axis=-1) * (1.0 - jnp.eye(n)))
   return ae, ee, r_ae, r_ee[..., None]