Let $D_1=\{|z-a|<r_1\}$,$D_2=\{|z-b|<r_2\}$,$D_3=\{|z-c|<r_3\}$ such that $D_1,D_2,D_3$ are disjoint.Let $f$ be an analytic function in $\mathbb{C}\diagdown(D_1\cup D_2\cup D_3)$.Prove that for some $A,B,C$,$f-\frac{A}{z-a}-\frac{B}{z-b}-\frac{C}{z-c}$ has a primitive in $\mathbb{C}\diagdown(D_1\cup D_2\cup D_3)$.

Instinctively I want find the explicit form of the primitive,the primitive of $f$ is the easy part,and I suppose the primitive of the fractions should take the logarithm form.But the it seems that $\mathbb{C}\diagdown(D_1\cup D_2\cup D_3)$ cannot be a analytic branch of logarithm.And I was stuck here.

**Update**Thanks to the hint by @Daniel Fischer,the statement that the primitive of $f$ is the easy part is not correct,since I just realize that $f$ is not analytic on the whole plane,then the integral over the closed loop in the domain is not necessarily zero.That been said,I think the key is to find a function of which the integral over closed loop is zero,then by the prove of Morera's theorem we can prove it's the primitive.

Any kind of help will be appreciated.

  • $\begingroup$ The centres of $D_2$ and $D_3$ should probably be $b$ and $c$, not $a$. $\endgroup$ May 18, 2014 at 14:14
  • $\begingroup$ Re the primitive of $f$ is the easy part, you may want to reconsider. Generally, $f$ has no primitive. $\endgroup$ May 18, 2014 at 14:16

1 Answer 1


Hint: $g \in \mathscr{O}(U)$ has a primitive if and only if

$$\int_\gamma g(z)\,dz = 0$$

for all closed (piecewise continuously differentiable) paths in $U$.


$$A = \frac{1}{2\pi i}\int\limits_{\lvert z-a\rvert = r_1+\varepsilon} f(z)\,dz,$$

where $\varepsilon > 0$ is small enough that the contour avoids the disks $D_2$ and $D_3$, and analogously $B,C$, we find that

$$\int_{\gamma} f(z) - \frac{A}{z-a} - \frac{B}{z-b} - \frac{C}{z-c}\,dz = 0\tag{1}$$

when $\gamma$ is one of the circles used above. Since every closed path in $U := \mathbb{C}\setminus (D_1\cup D_2\cup D_3)$ is homologous to a linear combination of these circles, Cauchy's integral theorem asserts that $(1)$ holds for all closed paths in $U$. Fixing an arbitrary $z_0\in U$,

$$F(z) = \int_{\gamma_z} f(\zeta)- \frac{A}{\zeta-a} - \frac{B}{\zeta-b}-\frac{C}{\zeta-c}\,d\zeta,$$

where $\gamma_z$ is an arbitrary path from $z_0$ to $z$ in $U$, is well-defined, and of course $F'(z) = f(z) - \frac{A}{z-a} - \frac{B}{z-b} - \frac{C}{z-c}$.

  • $\begingroup$ I can find a function of the form $k_1\log(z-a)+k_2\log(z-b)+k_3\log(z-c)$ satisfy zero integral on the closed paths,but how can we well-define such a logarithm analytically on the whole plane excluding three disks? $\endgroup$
    – Daniel S.
    May 18, 2014 at 14:28
  • $\begingroup$ In general, you can't have the logarithm part well-defined on the entire plane minus the disks. You'll have branch-cuts, but, and that is the point, the "quasi-primitive" of $f$ also has branch-cuts, and you can arrange it so that the cuts cancel. $\endgroup$ May 18, 2014 at 14:34
  • $\begingroup$ Thank you Daniel,you are an amazing teacher.Actually I can prove the existence of primitive perfectly fine following you hint,I'm curious about can I write the primitive explicitly? $\endgroup$
    – Daniel S.
    May 18, 2014 at 14:56
  • $\begingroup$ You can write the primitive "explicitly" as an integral. Given a particular $f$, you can often give a closed-form expression of the primitive, but since there are entire functions where you cannot give the primitive explicitly (except as a power series or using special functions), you cannot in general give the primitive explicitly here either. $\endgroup$ May 18, 2014 at 15:10

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.