How to define first order derivative of model output with respect to A SPECIFIC MODEL INPUT poinr

Hi Does anyone know if it is possible to define dy/dx at x = 0 in keras custom loss function? For solving 2nd order ODE with this boundary condition

I am not true if this is the correct definition: der = K.gradients(model.output, np.array([0]))[0] but this returns None, so it cannot work in the model training. Is there any way to define this?