5
$\begingroup$

There is an easy method to perform a partial fraction decomposition - described here, under the "Repeated Real Roots" title, for the coefficient A2.

The problem is - this method doesn't work in some cases, for example:

$$ \frac{1}{(2 + j \omega)^2 \, (4 + j \omega)} = \frac{A}{2 + j \omega} + \frac{B}{(2 + j \omega)^2} + \frac{C}{4 + j \omega} $$

The coefficients B,C get their proper value: $$ B = \frac{1}{4 + j \omega} \Big|_{\omega=2j} = \frac 12 $$ $$ C = \frac{1}{(2 + j \omega)^2} \Big|_{\omega=4j} = \frac 14 $$

But the coefficient A get a false value: $$ A = \frac{d}{d \omega} \left( \frac{1}{4 + j \omega} \right)\Big|_{\omega=2j} = \frac{-j}{4} $$ The true/proper value for A is $-\frac 14$.

I wonder what is the general formula for this kind of elegant/quick/easy trick to find the coefficients during partial fraction decomposition?

$\endgroup$
6
$\begingroup$

The partial fraction expansion is

$$\frac{1}{(2+j\omega)^2(4+j\omega)}=\frac{A}{2+j\omega}+\frac{B}{(2+j\omega)^2}+\frac{C}{4+j\omega}\tag 1$$

To find $C$, we multiply both sides of $(1)$ by $4+j\omega$ and take the limit as $j\omega \to -4$. Proceeding, we find

$$\begin{align} \lim_{j\omega \to -4}\frac{(4+j\omega)}{(2+j\omega)^2(4+j\omega)}&=\frac14\\\\ &=\lim_{j\omega \to -4}\left(\frac{A(4+j\omega)}{2+j\omega}+\frac{B(4+j\omega)}{(2+j\omega)^2}+\frac{C(4+j\omega)}{4+j\omega}\right)\\\\ &=C \end{align}$$

To find $B$ , we multiply both sides of $(1)$ by $(2+j\omega)^2$ and take the limit as $j\omega \to -2$. Proceeding, we find

$$\begin{align} \lim_{j\omega \to -2}\frac{(2+j\omega)^2}{(2+j\omega)^2(4+j\omega)}&=\frac12\\\\ &=\lim_{j\omega \to -2}\left(\frac{A(2+j\omega)^2}{2+j\omega}+\frac{B(2+j\omega)^2}{(2+j\omega)^2}+\frac{C(2+j\omega)^2}{4+j\omega}\right)\\\\ &=B \end{align}$$

To find $A$ , we multiply both sides of $(1)$ by $(2+j\omega)^2$, take a derivative with respect to $j\omega$, and take the limit as $j\omega \to -2$. Proceeding, we find

$$\begin{align} \lim_{j\omega \to -2}\frac{d}{d(j\omega)}\left(\frac{(2+j\omega)^2}{(2+j\omega)^2(4+j\omega)}\right)&=-\frac14\\\\ &=\lim_{j\omega \to -2}\frac{d}{d(j\omega)}\left(\frac{A(2+j\omega)^2}{2+j\omega}+\frac{B(2+j\omega)^2}{(2+j\omega)^2}+\frac{C(2+j\omega)^2}{4+j\omega}\right)\\\\ &=A \end{align}$$

$\endgroup$
  • $\begingroup$ Interesting, I haven't thought about that. Why taking the derivative with respect to $j \omega$ fixes the problem? $\endgroup$ – Dor Apr 21 '16 at 21:13
  • $\begingroup$ @Dor : at first we can find $B$ or $C$ as we wish, but not $A$ : the highest power of each roots first, only after the lower powers. (see my answer to see why) $\endgroup$ – reuns Apr 21 '16 at 21:14
  • $\begingroup$ As @user1952009 eloquently wrote, before differentiating, we have a first order term of $2+j\omega$ on $A$, while $B$ and $C$ have zeroth and second order terms respectively. $\endgroup$ – Mark Viola Apr 21 '16 at 21:20
  • $\begingroup$ Indeed, but why taking the derivative with respect to (only) $\omega$, instead of $j \omega$, isn't correct? $\endgroup$ – Dor Apr 21 '16 at 21:38
  • $\begingroup$ @dor You may take a derivative with respect to $\omega$ instead of $j\omega$. But then you have to worry about the pesky $j$'s floating around. $\endgroup$ – Mark Viola Apr 21 '16 at 21:52
4
$\begingroup$

suppose that you want to decompose $\frac{P(x)}{Q(x) (x-\rho)^n}$, with $Q(\rho) \ne 0$ :

$$\frac{P(x)}{Q(x) (x-\rho)^n} - \frac{a}{(x-\rho)^n} = \frac{P(x) - a Q(x) }{Q(x) (x-\rho)^n} = \frac{(x-\rho) S(x) + b }{Q(x) (x-\rho)^n} = \frac{S(x)}{Q(x) (x-\rho)^{n-1}} + \frac{b }{Q(x) (x-\rho)^n}$$

($S(x)$ and $b$ are obtained by Euclidean division of $P(x) - a Q(x)$ by $(x-\rho)$)

the target is to choose $a$ such that $b= 0$ : hence choosing $a$ such that the polynomial $P(x) - a Q(x)$ vanishes at $x = \rho$, which is always possible since $Q(\rho) \ne 0$ :

$$a = \frac{P(\rho)}{Q(\rho)}$$

hence, by induction we see that we can decompose any rational function :

$$\frac{P(x)}{\prod_i (x-\rho_i)^{e_i}} = R(x) + \sum_i \sum_{k=1}^{e_i} \frac{a_{i,k}}{(x-\rho_i)^{k}}$$

$\endgroup$
2
$\begingroup$

Here are the mechanics of the trick. Let's multiply your equation by $4+j\omega$:

$$ \frac{1}{(2 + j \omega)^2 \, } = \frac{(4 + j \omega)A}{2 + j \omega} + \frac{(4 + j \omega)B}{(2 + j \omega)^2} + C. $$

Now things nicely fall out if we look at $\omega = 4j$:

$$\frac{1}{(2-4)^2} = C.$$

But ... the left side is just the original expression with the $(4+j\omega)$ "covered up" (since we multiplied everything by $4+j\omega$). Evaluating this covered-up expression at the root we covered up gives the coefficient.

Things don't all cancel away with a repeated root. Multiply by $2 + j\omega$:

$$ \frac{1}{(2 + j \omega) \, (4 + j \omega)} = A + \frac{B}{(2 + j \omega)} + \frac{(2 + j \omega)C}{4 + j \omega}. $$

There's still the $2 + j \omega$ in the denominator under $B$ so the "trick" fails and we have to find it some other way.

$\endgroup$
1
$\begingroup$

Suppose $f(z)$ is a rational function with poles at $z = z_n$ in the complex plane. Then consider the Laurent expansion of $f(z)$ around each pole (a Laurent expansion is the generalization of the Taylor expansion where you now include negative powers). Let $g_n(z)$ denote the sum of all the singular terms of the expansion of $f(z)$ around $z = z_n$. Then consider the function $h(z)$ defined as:

$$h(z) = f(z) - \sum_n g_n(z)$$

Clearly $h(z)$ is then also a rational function, but since we've subtracted all the singularities from f(z), the function $h(z)$ only has removable singularities, so, it is in fact a polynomial. If $f(z)$ has a numerator with a lower degree than its denominator, then we see that $h(z)$ tends t zero at infinity, which means that $h(z)$ is actually zero everywhere. In that case we thus have:

$$f(z) = \sum_n g_n(z)$$

If the degree of the numerator is not smaller than the degree of the denominator, then you can still use this formula if you interpret the point at infinity as a singularity. You then include the expansion around infinity where you regard singular terms as terms with positive powers of $z$.

In this particular case we may cut some corners and obtain the partial fraction expansion with hardly any computations. We have:

$$f(z) = \frac{1}{(2+z)^2 (4+z)}$$

Obviously, the coefficient of $\frac{1}{z+4}$ is $\lim_{z\to -4}(z+4)f(z) = \frac{1}{4}$. To find the part of the partial fraction expansion involving powers of $\frac{1}{z+2}$, observe that $f(z)$ for large $z$ tends to zero as $\frac{1}{z^3}$. This means that the $\frac{1}{z+2}$ term must have a coefficient of $-\frac{1}{4}$. And the coefficient of $\frac{1}{(z+2)^2}$ is obviously given by $\lim_{z\to -2}(z+2)^2f(z) = \frac{1}{2}$

So. without having had to do much work, we have obtained the result:

$$f(z) = \frac{1}{2} \frac{1}{(z+2)^2} - \frac{1}{4} \frac{1}{z+2} + \frac{1}{4}\frac{1}{z+4}$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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