diff --git a/lineax/_solve.py b/lineax/_solve.py
index c339d5d..9b246f8 100644
--- a/lineax/_solve.py
+++ b/lineax/_solve.py
@@ -256,7 +256,7 @@ def _keep_undefined(v, ct):
         return None
 
 
-@eqxi.filter_primitive_transpose
+@eqxi.filter_primitive_transpose(materialise_zeros=True)  # pyright: ignore
 def _linear_solve_transpose(inputs, cts_out):
     cts_solution, _, _ = cts_out
     operator, state, vector, options, solver, _ = inputs
diff --git a/lineax/_solver/cg.py b/lineax/_solver/cg.py
index 573737a..b536f2f 100644
--- a/lineax/_solver/cg.py
+++ b/lineax/_solver/cg.py
@@ -124,15 +124,15 @@ def mv(vector: PyTree) -> PyTree:
             mv = operator.mv
         preconditioner, y0 = preconditioner_and_y0(operator, vector, options)
         leaves, _ = jtu.tree_flatten(vector)
+        size = sum(leaf.size for leaf in leaves)
         if self.max_steps is None:
-            size = sum(leaf.size for leaf in leaves)
             max_steps = 10 * size  # Copied from SciPy!
         else:
             max_steps = self.max_steps
         r0 = (vector**ω - mv(y0) ** ω).ω
         p0 = preconditioner.mv(r0)
         gamma0 = tree_dot(r0, p0)
-        rcond = resolve_rcond(None, vector.size, vector.size, vector.dtype)
+        rcond = resolve_rcond(None, size, size, jnp.result_type(*leaves))
         initial_value = (
             ω(y0).call(lambda x: jnp.full_like(x, jnp.inf)).ω,
             y0,