Why do high-frequency dynamics quickly go away in a step response?

As we know, a step input hits all the frequencies of a dynamical system. However, my professor told me today that the high-frequency response is only present for a short time at the very start, and then only the low-frequency parts remain for the rest of the step response. Here's an example:

The first image shows two identified systems - $P_1(j\omega)$ and $P_2(j\omega)$ - of the actual system $P(j\omega)$. As you see, $P_1(j\omega)$ matches closely at low frequencies while $P_2(j\omega)$ matches closely at high frequencies.

When we apply a step response, you can see that $P_2(j\omega)$ responds almost like the real system $P(j\omega)$ at the very start (what my professor says is the part where high frequencies are "activated") while $P_1(j\omega)$ is quite accurate for the rest "slow dynamics" part of the response - where my professor says the "low frequencies" remain.

Question : what is it that makes the high frequencies be present only at the start and dissipate away quickly as time goes by, while the low frequencies persist throughout the step response, decaying more slowly?

Thanks a lot for your help!

• First of all, your identified system number 2 seems to have poor estimation accuracy of the static gain. I don't know how you judge the identification accuracy in this regard. Sep 17 '15 at 10:30

To make this all more precise, just look at the expansion of a system's time-response in complex-exponential terms of the type $e^{(a+jb)t}$.