How to calculate the Jacobian of odometry error?

[Junkai-Liu]

I don't know how to get the jacobian file (such as intel_jac.txt)from the observation of odometry as well as estimate of pose and covariance matrix of estimate uncertainty(such as the format of g2o file ,if u familier with it).
To illustrate the issue ,in the case of pose-graph SLAM,denoted motion model noisy w~R(0,R^(-1)).In order to simplify the problem ,assumpt there only two poses of robots, in the case of SE(2), P0(x0,y0,theta0),P1(x1,y1,theta1),then the estimate state is Pi=(x0,y0,theta0,x1,y1,theta1),then the odometry prediction is h=[delta_x,delta_y,delta_theta]^{T} =[x1-x0,y1-y0,theta1-theta0],then
dh/dx=R^{-T/2}[-1,0,0,1,0,0; 0,-1,0,0,1,0; 0,0,-1,0,0,1], its the result derived from theory.Obviously,it is not match your dataset (I assume that the estimated covariance matrix is invariant.).I wonder whether I understand something wrong or not.Looking forward to your reply.Thank u very much