Questions about the loss function

[HanatoK]

Is this an implementation of SRVs reported in https://doi.org/10.1063/1.5092521? If so, as I read in the paper the loss function in Figure 1 only includes C00 ($\mathrm{E}\left[f_{j}(x_t)f_{i}(x_t)\right]$)
and C01 ($\mathrm{E}\left[f_{j}(x_t)f_{i}(x_{t+\tau})\right]$), but in the code (

srv/hde/hde.py

Lines 301 to 318 in 6887ef4

	C10 = 1/(N - 1)*K.dot(K.transpose(z_tt), z_t0)
	C11 = 1/(N - 1)*K.dot(K.transpose(z_tt), z_tt)
	if not self.reversible:
	vamp_matrix = K.dot(K.dot(_inv(C00, ret_sqrt=True), C01), _inv(C11, ret_sqrt=True))
	vamp_score = tf.norm(vamp_matrix)
	return -1.0 - tf.square(vamp_score)
	else:
	C0 = 0.5*(C00 + C11)
	C1 = 0.5*(C01 + C10)
	L = tf.cholesky(C0)
	Linv = tf.matrix_inverse(L)
	A = K.dot(K.dot(Linv, C1), K.transpose(Linv))
	lambdas, _ = tf.self_adjoint_eig(A)
	return -1.0 - K.sum(self.weights*lambdas**2)

)
why are C10 and C11 involved?