Update MLIAP JAX example to use jax.grad

examples/mliap/jax/mliap_unified_jax.py

@@ -5,10 +5,30 @@ import jax.numpy as jnp
 from jax import jit
 from functools import partial
 
+#               /- ensure epsilon and sigma are treated as compile-time constants
+@partial(jit, static_argnums=(0, 1))
+def lj_potential(epsilon: float, sigma: float, rij):
+    # a pure function we can differentiate:
+    def _tot_e(rij):
+        r2inv = 1.0 / jnp.sum(rij ** 2, axis=1)
+        r6inv = r2inv * r2inv * r2inv
+
+        lj1 = 4.0 * epsilon * sigma**12
+        lj2 = 4.0 * epsilon * sigma**6
+
+        eij = r6inv * (lj1 * r6inv - lj2)
+        return jnp.sum(eij)
+    #                   /-  construct a function computing _tot_e and its derivative
+    tot_e, fij = jax.value_and_grad(_tot_e)(rij)
+    return tot_e, fij
+
 
 class MLIAPUnifiedJAX(MLIAPUnified):
     """Test implementation for MLIAPUnified."""
 
+    epsilon: float
+    sigma: float
+
     def __init__(self, element_types, epsilon=1.0, sigma=1.0, rcutfac=1.25):
         # ARGS: interface, element_types, ndescriptors, nparams, rcutfac
         super().__init__(None, element_types, 1, 3, rcutfac)
@@ -39,23 +59,7 @@ class MLIAPUnifiedJAX(MLIAPUnified):
         #       This might account for ~2-3x slowdown compared to original LJ.
         rij = np.pad(rij, ((0,npad), (0,0)), 'constant')
 
-        eij, fij = self.compute_pair_ef(rij)
+        e_tot, fij = lj_potential(rij)
 
-        data.update_pair_energy(np.array(np.double(eij)))
+        data.energy = e_tot.item()
-        data.update_pair_forces(np.array(np.double(fij)))
+        data.update_pair_forces(np.array(fij, dtype=np.float64))
-
-    #@jax.jit # <-- This will error! See https://github.com/google/jax/issues/1251
-    # @partial takes a function (e.g. jax.jit) as an arg.
-    @partial(jax.jit, static_argnums=(0,))
-    def compute_pair_ef(self, rij):
-
-        r2inv = 1.0 / jnp.sum(rij ** 2, axis=1)
-        r6inv = r2inv * r2inv * r2inv
-
-        lj1 = 4.0 * self.epsilon * self.sigma**12
-        lj2 = 4.0 * self.epsilon * self.sigma**6
-
-        eij = r6inv * (lj1 * r6inv - lj2)
-        fij = r6inv * (3.0 * lj2 - 6.0 * lj2 * r6inv) * r2inv
-        fij = fij[:, jnp.newaxis] * rij
-        return eij, fij