In computeEdgeSE3Gradient, why the jacobians of dte/dtj is related to Rj

[issacfrog]

Hello, g2o team.I’m learning about g2o optimizer.I had some questions about computeEdgeSE3Gradient function.
The code is as follow:

const Isometry3 A=Z.inverse();
const Isometry3 B=Xi.inverse()Xj;
E=AB; // [Rz^-1 * Ri^-1 * Rj | Rz^-1 * Ri^-1 * tj - Rz^-1 ti - tz]*
Isometry3::ConstLinearPart Re = extractRotation(E);
Isometry3::ConstLinearPart Ra = extractRotation(A);
Isometry3::ConstLinearPart Rb = extractRotation(B);
Isometry3::ConstTranslationPart tb = B.translation();

Eigen::Matrix<double, 3, 9, Eigen::ColMajor> dq_dR;
compute_dq_dR(dq_dR, Re(0, 0), Re(1, 0), Re(2, 0), Re(0, 1), Re(1, 1),
Re(2, 1), Re(0, 2), Re(1, 2), Re(2, 2));
Ji.setZero();
Jj.setZero();
// dte/dti
Ji.template block<3, 3>(0, 0) = -Ra; // -Ra = - Rz^-1
// dte/dtj
Jj.template block<3, 3>(0, 0) = Re; // Re = Rz^-1 * Ri^-1 * Rj or Rz^-1 * Ri^-1
Is there something wrong with my formula ?