A local construction can fail globally
Near a nonzero complex number, a branch of the logarithm can be chosen. It is tempting to continue such a branch around any domain and assume that the values agree when we return to the starting point. On an annulus surrounding zero they do not: one circuit changes a continuously followed argument by . A logarithm is therefore a question about both the function and the loops in its domain.
Let be a connected open set and let be holomorphic and zero-free on .
Definition 3.4.1
(Holomorphic logarithm and holomorphic root)
- A holomorphic logarithm of is a holomorphic on with .
- For a positive integer , a holomorphic th root of is a holomorphic on with .
Because is zero-free, any such is also zero-free. Differentiating a proposed logarithm or root gives
These identities provide necessary tests, but the reverse direction requires a construction and a check that the constructed value does not depend on the chosen path.
Periods of the logarithmic derivative
Proposition 3.4.2
(Periods of the logarithmic derivative are integers)
For any piecewise closed curve in ,
Proof
(Following a logarithm along the image curve)
Write , a nonzero closed piecewise curve. On each smooth piece define , joining the integrals across corners. Then is constant. Since returns to its initial value, , so is an integer multiple of . This integer is the winding number of around zero.
Common mistake
(Counting the winding of the wrong curve)
The integer is the winding number of the image curve about , not the winding number of itself. For and the unit circle , the curve winds once about , but , because runs around the unit circle twice.
Theorem 3.4.3
(Logarithm criterion)
A zero-free holomorphic has a holomorphic logarithm on if and only if
Proof
(Integrating the logarithmic derivative)
If , then and every closed integral vanishes. Conversely fix . Open connected subsets of the plane are polygonally connected, so piecewise paths exist between their points; define
The hypothesis makes the integral path-independent, since two paths followed in opposite directions form a closed curve. In a disk about any , the straight segment to a nearby lies in the disk, so the difference quotient of the integral tends to ; hence . Hence , and connectedness makes a nonzero constant . Choose any complex logarithm of that single number. Then is holomorphic and . The choice of affects the answer by an integer multiple of ; it does not affect existence.
The construction is worth reading once more with the role of each hypothesis in view.
Proof X-ray
The loop test closes the path-integral construction
The local derivative is available everywhere because has no zeros. To integrate it globally, compare two paths to the same point: their difference is a closed loop. Vanishing of all loop integrals gives one well-defined primitive. The derivative of then vanishes, leaving only a constant normalization. Each assumption has a role: connectedness fixes one constant, and zero-freeness makes the logarithmic derivative holomorphic.
Roots need less than logarithms
Theorem 3.4.4
(Root criterion)
A zero-free holomorphic has a holomorphic th root on if and only if every period of is an integer multiple of .
Proof
(Dividing the primitive by )
If , then , and the preceding integer-period result gives the condition. Conversely choose and a number with . Set
By the assumed period condition, changing paths changes the exponent by an integer multiple of , so its exponential, and hence , is single-valued. Locally it is the exponential of a holomorphic primitive and hence holomorphic. Its logarithmic derivative is , so . The ratio equals one at and is therefore one throughout . For the condition is automatic, as it must be; for general , a root can exist even when no logarithm exists.
Examples and a counterexample
Worked example 3.4.5
(The punctured plane)
For on , the unit circle has . Thus has no holomorphic logarithm there. For it has no holomorphic th root there either, because the period is not a multiple of . This conclusion concerns the whole punctured plane; local logarithms and roots still exist around each individual nonzero point.
Worked example 3.4.6
(A root without a logarithm)
On the same domain has the holomorphic square root , while , so it has no global logarithm. Failure of a logarithm does not imply failure of every root. The root criterion agrees: the period is a multiple of , but not of , so has no holomorphic cube root there.
Worked example 3.4.7
(A slit removes the obstruction)
On , write with . The principal branch is holomorphic with derivative and exponential . Consequently is holomorphic there. The cut changes the domain: a closed loop winding once around zero cannot lie entirely in this slit plane.
Common mistake
(Using the principal branch across the cut)
The principal logarithm is a logarithm of on the slit plane only. Approaching from above gives , from below : a jump of across the negative real axis. It therefore cannot be used as a holomorphic logarithm on an annulus around , which is consistent with the period computation above.
Counterexample mode
Local branches need not patch around a loop
Every point of has a small disk on which admits a holomorphic logarithm, yet the whole domain has none. The obstruction is the nonzero integral , not a failure of differentiability near any individual point.
Practice and solutions
Quick check
(On , does have a holomorphic cube root? A holomorphic logarithm?)
Compute the unit-circle period of .
Solution · Answer
is a cube root; in agreement with the root criterion, the period of is a multiple of . Since , no logarithm exists.
Quick check
(On the punctured plane, can hold for a holomorphic ?)
Apply the root criterion with .
Solution · Answer
No: the period of is , not an integer multiple of ; a square root would force that divisibility.
Quick check
(If has two holomorphic logarithms on a connected domain, how are they related?)
Exponentiate the difference.
Solution · Answer
, so takes values in the discrete set . Continuity and connectedness make it a constant with .
Quick check
(Why does every zero-free holomorphic function on a simply connected domain have a holomorphic logarithm?)
Check the hypothesis of the logarithm criterion.
Solution · Answer
The simply connected-domain form of Cauchy's theorem gives zero integrals of the holomorphic function around every closed curve. The criterion therefore gives a logarithm.