Docstring fixes for PSD kernels.

PiperOrigin-RevId: 474668459

Author: emilyfertig

tensorflow_probability/python/math/psd_kernels/internal/util.py

@@ -204,7 +204,7 @@ def pairwise_square_distance_matrix(x1, x2, feature_ndims):
 
   The latter term can be written as a matmul between `x1` and `x2`.
   This reduces the memory from the naive approach of computing the
-  squared difference of `x1` and `x2` by a factor of `(prod_k D_k) ** 2`.
+  squared difference of `x1` and `x2` by a factor of `prod_k D_k`.
   This is at the cost of the computation being more numerically unstable.
 
   Args:

tensorflow_probability/python/math/psd_kernels/parabolic.py

@@ -30,7 +30,7 @@ class Parabolic(psd_kernel.AutoCompositeTensorPsdKernel):
   """The Parabolic kernel.
 
   ```none
-  k(x, y) = 3 / 4 * amplitude * max(0, 1 - (||x_k - y_k|| / length_scale**2)
+  k(x, y) = 3 / 4 * amplitude * max(0, 1 - (||x_k - y_k|| / length_scale)**2)
   ```
 
   where the double-bars represent vector length (ie, Euclidean, or L2 norm).