Polynomial root multiplicity and residue theorem

397 Views Asked by At

Let $p(x) = \prod_{i} \left( x - \lambda_i \right)^{m_i}$ be a polynomial where the roots $\lambda_i$ may be complex but $p(x)$ has real coefficients (so complex roots must come in conjugate pairs).

It is know that the logarithmic derivative of $p(x)$ has the form

$$ \frac{p'(x)}{p(x)} = \sum_{i} \frac{m_i}{x - \lambda_i}. $$

Let $\gamma$ be a closed contour in the complex plane that contains the complex root $\lambda_1$ and does not contain any other root. Applying the residue theorem then we have

$$ \frac{1}{2 \pi i}\oint_\gamma \frac{p'(x)}{p(x)} dx = m_1. $$

This is because $m_1$ is the residue of $\frac{p'(x)}{p(x)}$ at $\lambda_1$. Indeed:

$$ \lim_{x\to\lambda_1} \left(x - \lambda_1 \right) \frac{p'(x)}{p(x)} = m_1. $$

First question: is my understanding up to here correct?

On the other hand, I am trying to solve this integral by hand as follows. Parameterize $\gamma$ as $x(t) = \lambda_1 + \epsilon e^{i t}$ for $t \in [0, 2\pi]$m, where $\epsilon$ is small enough so that $\gamma$ only contains $\lambda_1$ and no other root. Then we have

$$ \oint_\gamma \frac{p'(x)}{p(x)} dx = \int_0^{2 \pi} \frac{p'(\lambda_1 + \epsilon e^{i \pi t})}{p(\lambda_1 + \epsilon e^{i t})} (\epsilon i e^{i t}) dt = \ln ( p(\lambda_1 + \epsilon e^{i 2 \pi})) - \ln ( p(\lambda_1 + \epsilon e^{i 0})) = \ln (p(\lambda_1 + \epsilon )) - \ln (p(\lambda_1 + \epsilon )) = 0. $$

Can anyone please correct this last integral? I am pretty sure the first part of my understanding is correct. What is the problem with the way I'm solving the integral? TIA.

1

There are 1 best solutions below

3
On BEST ANSWER

Yes, your understanding up to the point where you get that the residue is $m_1$ is correct.

Concerning the rest, the problem lies in using an undefined function, which is $\ln$. There is no holomorphic function $\ln\colon\Bbb C\setminus\{0\}\longrightarrow\Bbb C$ such that $(\forall z\in\Bbb C\setminus\{0\}):\ln'(z)=\frac1z$. If there was, we would have $\int_{|z|=1}\frac1z\,\mathrm dz=0$, by the same argument as yours, but, in fact, $\int_{|z|=1}\frac1z\,\mathrm dz=2\pi i$.