Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

It was pretty hard finding a short and precise title. Here is my problem:

The equation $$\bigg|\int_\gamma f(z)\text{d}z\bigg|\le\int_\gamma\big|f(z)||\text{d}z|$$ holds if $f$ is integrable (where $\gamma$ is a differentiable path $[0,1]\rightarrow\mathbb{C}$). However, in what cases does "$=$" apply? Some cases are clear; for example if $f(z)=$const, $\forall z\in {\rm Img}(\gamma)$.

Our second idea was that both sides are equal if (or even iff?) $f(z)$ stays within one quadrant of the complex plane for all points in ${\rm Img}(\gamma)$.

So basically, the question is, when does "$=$" apply and is our second idea correct? Use "if" or "iff" (which is if and only if). Is there a nice proof to whatever answer there is? (No need to print the full proof; the basic idea suffices)

EDIT: instead of using the path definition of the integral, here is the corrected version: $$\bigg|\int_0^1f(\gamma(t))\gamma'(t)\text{d}t\bigg|\le\int_0^1|f(\gamma(t))||\gamma'(t)|dt.$$

share|cite|improve this question
I think on the right hand side of your inequality $dz$ should be replaced by $|dz|$. – 23rd Dec 14 '12 at 17:55
this was the inequation we proved in our analysis reading; how id $|\text{d}z|$ defined? I will edit my question. – CBenni Dec 14 '12 at 17:58
$|dz|$ denotes the infinitesimal arc length, i.e. if $\gamma$ is parameterized as $z=\gamma(t)$, then $dz=\gamma'(t)dt$ and $|dz|=|\gamma'(t)|dt$ – 23rd Dec 14 '12 at 18:07
@richard ok, that makes sense to me ;) Back to the question, any idea? – CBenni Dec 14 '12 at 18:08
I still need you to clarify in what sense $f$ is integrable? – 23rd Dec 14 '12 at 18:28
up vote 3 down vote accepted

It suffices to consider the following reduced question: if $g:[0,1]\to\mathbb{C}$ is Riemann/Lebesgue integrable, in the following inequality $$\left|\int_0^1g(t)dt\right|\le \int_0^1|g(t)|dt,\tag{1}$$ when does "$=$" holds?

To answer this question, let us recall how to deduce $(1)$. Since if $g$ is Riemann integrable on $[0,1]$, it is automatically Lebesgue integrable, so let us focus on the Lebesgue integrable case. Denote $I=\int_0^1g(t)dt$. If $I=0$, then $(1)$ is trivially true, and "$=$" holds if and only if $g=0$ almost everywhere. If $I\ne 0$, then there exists a unique $\alpha\in\mathbb{C}$ with $|\alpha|=1$, such that $I=|I|\alpha$. Therefore, $$|I|=\overline{\alpha}I=\mathrm{Re}(\overline{\alpha}I)=\mathrm{Re}\int_0^1\overline{\alpha}g(t)dt=\int_0^1\mathrm{Re}(\overline{\alpha}g(t))dt\le\int_0^1|g(t)|dt.$$ Note that for the last inequaltiy, "$=$" holds if and only if $\mathrm{Re}(\overline{\alpha}g(t))=|g(t)|$ for almost every $t\in[0,1]$, i.e. $g(t)=\alpha|g(t)|$ for almost every $t\in[0,1]$. To sum up, "$=$" in $(1)$ holds if and only if there exists $\alpha\in\mathbb{C}$ with $|\alpha|=1$, such that $g(t)=\alpha|g(t)|$ for almost every $t\in[0,1]$. To go back to your original question, simply let $g(t)=f(\gamma(t))\gamma'(t)$.

share|cite|improve this answer
I finally understood your proof ^.^ Heres the accept. – CBenni Dec 16 '12 at 9:13
@CBenni: Thank you! Actually this is a classical proof which can be found in standard textbooks of complex analysis. For example, I just checked that it appeared in section 1.1, chapter 4 of Ahlfors' complex analysis. – 23rd Dec 16 '12 at 13:07
Well that amazes me to be honest. We asked both our professor as well as our assistant and even dozens of fellow students, noone knew an answer right away... – CBenni Dec 16 '12 at 13:16
@CBenni: Really? Have you heard Minkowski inequality? In fact the inequality in your question is a very special case of the Minkowski's integral inequality. – 23rd Dec 16 '12 at 13:26
I havent (Until I read wisefools answer), but thank you nonetheless ;) – CBenni Dec 16 '12 at 13:29

If you write $$\bigg|\int_0^1f(\gamma(t))\gamma'(t)\text{d}t\bigg|\le\int_0^1|f(\gamma(t))||\gamma'(t)|dt$$ then it should be clear that you are asking this other question: take $a,b:[0,1]\to\mathbb{R}$ and write $$A=\int_0^1a(t)dt\qquad B=\int_{0}^1b(t)dt\;,$$ when does one have $$\sqrt{A^2+B^2}=\int_0^1(a^2(t)+b^2(t))^{1/2}dt$$ ? By squaring you get $$A^2+B^2= \left(\int_{0}^{1}(a^2(t)+b^2(t))^{1/2}dt\right)^2=\|a^2+b^2\|_{1/2}$$ and then $$A^2+B^2=\left(\int_0^1 a(t)dt\right)^2+\left(\int_{0}^1b(t)dt\right)^2=\left(\int_0^1 \sqrt{a^2(t)}dt\right)^2+\left(\int_0^1 \sqrt{b^2(t)}dt\right)^2=\left(\int_{0}^{1}(a^2(t)+b^2(t))^{1/2}dt\right)^2$$ because $$A^2+B^2\le\|a^2\|_{1/2}+\|b^2\|_{1/2}\le\|a^2+b^2\|_{1/2}$$ which is the reverse Minkowski inequality for $p=1/2$ (Minkowski and Holder hold for $0<p<1$ with the opposite inequality). The only way to have equality in the first one is $a=\pm|a|$, $b=\pm|b|$ to hold for a.e. $t$;for the second, like in the usual Minkowsky inequality, is to have $a^2=\lambda^2 b^2$ for a fixed real $\lambda$, that is $a+ib=(\lambda + i)b$.

Now, in your setting this means that $f(\gamma(t))\gamma'(t)$ has to be of the form $e^{i\alpha}\phi(t)$, where $\alpha\in\mathbb{R}$ is fixed and $\phi:[0,1]\to\mathbb{R^+}$, i.e. $$f(\gamma(t))=\frac{e^{i\alpha}\phi(t)}{\gamma'(t)}\;.$$

If, wlog, you take $\gamma$ a $\mathcal{C}^1$ curve, parametrized with respect to arc length, then $|\gamma'(t)|=1$ and $\gamma'(t)=e^{i\beta(t)}$; then you have $$f(\gamma(t))=e^{i\alpha-i\beta(t)}\phi(t)$$ which, upon rescaling $f$ by a constant factor, can be reduced to $$f(\gamma(t))=e^{-i\beta(t)}\phi(t)$$ which is not as neat as before (errors always improve results!) but yields nonetheless some link with curvature: $$\frac{df}{dz}(\gamma(t))e^{i\beta(t)}=f'(\gamma(t))\gamma'(t)=\frac{d}{dt}f(\gamma(t))=-i\beta'(t)e^{-i\beta(t)}\phi(t)+e^{-i\beta(t)}\phi'(t)$$ that is $$\frac{df}{dz}(\gamma(t))=e^{-2i\beta(t)}(-i\beta'(t)\phi(t)+\phi'(t))$$ and $\beta'(t)$ is related to the curvature of $\gamma$.

share|cite|improve this answer
@CBenni: The correct form of the sixth line should be $\sqrt{A^2+B^2}=\int_0^1\sqrt{a^2(t)+b^2(t)}dt$. – 23rd Dec 14 '12 at 21:12
Whops, you are right, I'll correct it (and what follows, obviously). – wisefool Dec 14 '12 at 21:34

For simplicity's sake define $$ g(t)=f(\gamma(t))\gamma'(t)\tag{1} $$ Note that $$ \int_0^1|g(t)|\,\mathrm{d}t =\sup_{|u(z)|=1}\mathrm{Re}\left(\int_0^1g(t)u(t)\,\mathrm{d}t\right)\tag{2} $$ and $$ \left|\,\int_0^1g(t)\,\mathrm{d}t\,\right| =\sup_{\substack{|u(z)|=1\\u\text{ constant}}}\mathrm{Re}\left(\int_0^1g(t)u(t)\,\mathrm{d}t\right)\tag{3} $$ By the principle that the sup over a larger set is larger, $(2)\ge(3)$.

Suppose that $(3)\ge(2)$. Then, by the compactness of the unit circle, there is a $u_0\in\partial D$ so that for any function $u:[0,1]\mapsto\partial D$, we have $$ \mathrm{Re}\left(\int_0^1g(t)u(t)\,\mathrm{d}t\right)\le\mathrm{Re}\left(\int_0^1g(t)u_0\,\mathrm{d}t\right)\tag{4} $$ Set $$ u(t)=\left\{\begin{array}{} \frac{|g(t)|}{g(t)}&\text{when }g(t)\ne0\\ 1&\text{when }g(t)=0 \end{array}\right.\tag{5} $$ Then we get from $(4)$ that $$ \begin{align} \int_0^1|g(t)|\,\mathrm{Re}\left(1-\frac1{u(t)}u_0\,\right)\,\mathrm{d}t &=\int_0^1\,\mathrm{Re}\left(\,g(t)u(t)-g(t)u_0\,\right)\,\mathrm{d}t\\ &\le0\tag{6} \end{align} $$ However, since $\left|\frac1{u(t)}\right|=1$ and $|u_0|=1$, we have $\mathrm{Re}\left(1-\frac1{u(t)}u_0\,\right)\ge0$. Since $(6)$ represents the integral of two non-negative functions, yet is non-positive, we get that for almost all $t\in[0,1]$, either $g(t)=0$ or $u(t)=u_0$. Thus, $$ g(t)=\frac{|g(t)|}{u_0}\tag{7} $$ Therefore, equality can hold only when $g(t)$ has the same argument for all $t\in[0,1]$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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