I just discovered the Kramers-Kronig relation and am trying to apply it to a simple damped oscillator of the form subjected to an impulse at $t=0$, which is a causal system:

$$m\ddot x + c\dot x + k x = \delta(t).$$

In the time-domain, the response can be decomposed in odd/even parts as $x(t) = \operatorname{sign}(t)h_0(t) + h_0(t)$. The Kramers-Kronig relation implies that because the signal is causal, the real part of the Fourier transform of a solution $x(t)$ is equal to the Hilbert transform of the imaginary part of the Fourier transform of $x(t)$. That is what I am trying to illustrate here (and I am afraid that it would lead to a constraint between $m$, $c$ and $k$, which would probably suggest a misunderstanding on my side...).

The Fourier transform of the ODE gives $$ (-m \omega^2 +k + ic\omega )\hat x = 1$$


$$\hat x = \dfrac{1}{-m \omega^2 +k + ic\omega } = \dfrac{k-m\omega^2}{(k-m\omega^2)^2 + (c\omega )^2} + i \dfrac{-c\omega}{(k-m\omega^2)^2 + (c\omega )^2} $$

If I am not mistaken, the KK relation would imply that the Hilbert transform of $\dfrac{-c\omega}{(k-m\omega^2)^2 + (c\omega )^2} $ is $\dfrac{k-m\omega^2}{(k-m\omega^2)^2 + (c\omega )^2}$, ie

$$\dfrac{1}{\pi}\int_{-\infty}^\infty -\dfrac{1}{u-\omega}\dfrac{c\omega}{(k-m\omega^2)^2 + (c\omega )^2} d\omega \stackrel{?}{=} \dfrac{k-m u^2}{(k-m u^2)^2 + (cu )^2}$$ which does not hold (take for instance $u=m=c=k=1$, the integral does not converge). Edit Actually, it does hold for $m=c=k=1$, in the sense of the principal value (see hint and answer) ! I was not able to verify for arbitrary $m,c,k$.

Mathematica code for whoever is interested:

m = k = c = 1;
LHS  = Assuming[Element[u, Reals], 1/Pi*Integrate[-1/(u - w)*c*
      w/((k - m*w^2)^2 + (c^2*w^2)), {w, -Infinity, Infinity}, 
     PrincipalValue -> True]];
RHS = (k - m*u^2)/((k - m*u^2)^2 + c^2*u^2);
LHS == RHS // Simplify
  • 2
    $\begingroup$ Quick hint : Hilbert transform of a function $f(x)$ is $\frac{1}{\pi}P.V.\int_{-\infty}^{+\infty}dx'\frac{f(x')}{x-x'}$. Your integral certainly converges in the sense of a principal value. $\endgroup$
    – Sunyam
    Commented Mar 1, 2019 at 21:36
  • $\begingroup$ @Sunyam Out of curiosity, how do you know it converges in the PV sense? $\endgroup$
    – anderstood
    Commented Mar 7, 2019 at 19:13
  • $\begingroup$ Only singularity your integral (without interpreting it in the sense of a principal value) along real axis is at $\omega=u$ which disappears when interpreted as a principal value (calculated by using semi circular contour in upper/lower plane + real axis with a dent bypassing the singular point). $\endgroup$
    – Sunyam
    Commented Mar 7, 2019 at 20:32

1 Answer 1


This work!

$$x=\frac{1}{-m{\omega}^{2}+k+ic\omega}=\Re{(x(\omega))}+i\,\Im{(x(\omega))}$$ with: $$\Re{(x(\omega))}={\frac {-m{\omega}^{2}+k}{ \left( -m{\omega}^{2}+k \right) ^{2}+{c}^{2 }{\omega}^{2}}}\tag 1$$ and $$\Im{(x(\omega))}=-{\frac {c\omega}{ \left( -m{\omega}^{2}+k \right) ^{2}+{c}^{2}{\omega }^{2}}} \tag 2$$

Kramers-Kronig relation :

$$I_R=-\frac{2}{\pi}\int_0^\inf \frac{\omega\,\Im{(x(\omega))}}{\omega^2-s^2}\,d\omega$$

for $m=1\,,k=1\,,c=1$

we get:


this is the value of equation (1) for $\Re{(x(s))|_{m=1\,,k=1\,,c=1}}$

$$I_I=\frac{2}{\pi}\int_0^\inf \frac{s\,\Re{(x(\omega))}}{\omega^2-s^2}\,d\omega$$

we get:


this is the value of equation (2) for $\Im{(x(s))|_{m=1\,,k=1\,,c=1}}$

to get the right answer you must take the Cauchy-Principal -Value of the integral result :

$$I= \text{Cauchy-Principal -Value}\,\left[\int_0^\inf (...)\,d\omega\right]$$

I did the calculation with Maple symbolic program

enter image description here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.