What does infinity reveal about an entire map?
An entire function has no finite singularities, but it can behave very differently far from the origin. A polynomial grows like a power in every sufficiently large direction; grows along the positive real axis and tends to zero along the negative axis. To study infinity using ordinary local tools, replace by . The behavior of as becomes the isolated singularity of at .
Definition 1.7.1
(Removable singularities, poles and essential singularities)
Let be holomorphic on a punctured disk . The isolated singularity is
- removable if extends holomorphically across ;
- a pole if as ; then has a removable singularity with nonzero value at for exactly one integer , the order of the pole;
- essential if it is neither removable nor a pole.
A zero-free bounded holomorphic function on a punctured disk may extend with value zero at its center, as does on ; this distinction will matter in the proof below.
A density theorem near an essential singularity
Theorem 1.7.2
(Casorati–Weierstrass theorem)
Suppose is holomorphic on and has an essential singularity at . For every , the set is dense in .
Proof
(An omitted disk makes the reciprocal bounded)
Suppose otherwise. Then some open disk centered at is disjoint from the image of a punctured neighborhood of : for some ,
The function is holomorphic and bounded by there, so the removable-singularity theorem extends it holomorphically to . If , then extends holomorphically across . If , the extended is not identically zero (it is nonzero on the punctured disk) and has a zero of finite order; has a pole at , so has a pole there. Both possibilities contradict essentiality. Thus no such omitted disk exists, exactly the asserted density.
Proof X-ray
Why an omitted disk controls the singularity
The hypothesis is negated in its strongest usable form: not merely a missed value, but a missed open disk, which yields the uniform lower bound . That bound is the only input to the removable-singularity theorem, so the key move is passing to , which converts “stays away from ” into “bounded”. Omitting a single value would give a zero-free but no bound on , and the argument would stall.
The proof closes by a two-case split at the extended value . A nonzero value gives a removable singularity for , while a zero gives a pole; the finite order of that zero uses that is not identically zero, which holds because never vanishes on the punctured disk. An essential singularity permits neither case.
Density is weaker than surjectivity. The stronger great Picard theorem says much more, but is not needed for the argument here and is not proved in this unit.
Counterexample mode
Dense does not mean every value is attained
A plausible strengthening reads: near an essential singularity, takes every complex value. For , whose Laurent series has infinitely many negative powers, the singularity at is essential, yet for every because the exponential never vanishes. The value is omitted on every punctured disk. An omitted point has empty interior and is compatible with a dense image; an omitted open disk is the obstruction ruled out by the proof. The correct statement is density of the image.
Infinity: removable, pole, or essential
Proposition 1.7.3
(Entire functions classified by their singularity at infinity)
Let be entire and for . Then has a removable singularity at if and only if is constant; a pole of order if and only if is a polynomial of degree ; and an essential singularity if and only if is not a polynomial.
Proof
(Liouville and Cauchy estimates at infinity)
If has a removable singularity at zero, then is bounded outside a large disk and hence bounded on the whole plane; Liouville's theorem makes constant. If has a pole of order , then near , that is, for large . Cauchy's estimates on give for the Taylor coefficients of , and letting shows for ; so is a polynomial of degree at most . Conversely, a polynomial with , , gives , a pole of order exactly , and a constant gives a removable singularity. Since the three cases for are exclusive and exhaustive, the degree must equal , and any nonpolynomial entire function has an essential singularity at infinity.
The phrase “singularity at infinity” is shorthand for the isolated singularity of at zero; it does not assert that infinity is a point of the original complex plane.
Common mistake
(Tending to infinity characterizes polynomials)
It is tempting to think that every nonconstant entire function satisfies as . That is true for nonconstant polynomials and false for every other entire function: if , then as , which is a pole, so is a polynomial. For instance along the negative real axis.
Classifying injective entire functions
Theorem 1.7.4
(Injective entire functions are affine)
Every injective entire function has the form with . Conversely, each such map is injective and entire.
Proof
(Infinity is a pole, and the derivative never vanishes)
Injectivity makes nonconstant. Its image of the unit disk is a nonempty open set by the open mapping theorem. Injectivity also makes this open set disjoint from the image of the exterior . If infinity were essential, Casorati–Weierstrass applied to would make the exterior image dense in , contradicting its omission of a nonempty open set. Infinity is therefore not essential. It cannot be removable, because then would be constant; hence it is a pole and is a polynomial.
It remains to rule out degree greater than one. An injective holomorphic map has nonzero derivative: if the first nonzero term of the Taylor expansion of about had degree , choose a circle about so that has no zeros in the punctured closed disk and on the circle. Rouché's theorem then gives roots of inside it for sufficiently close to . None is , and all are simple because is nonzero there, so they are distinct preimages. Thus is a polynomial with no zero. By the fundamental theorem of algebra, a nonconstant complex polynomial has a zero, so must be a nonzero constant. Therefore with . Conversely, forces .
Common mistake
(A nonvanishing derivative does not give global injectivity)
The conclusion depends on the domain being the whole plane, and on injectivity on all of it. The exponential is locally conformal everywhere because its derivative never vanishes, but , so it is not globally injective. In the other direction, is injective on the right half-plane even though it is not injective on the whole plane: if then , and only keeps both points in the half-plane.
Worked examples
Worked example 1.7.5
(Every nonzero value of near , explicitly)
Fix and choose any logarithm of , so . The equation holds exactly when for some . For all but at most one the right side is nonzero, and
So every nonzero value is attained infinitely often in every punctured neighborhood of , which is far more than density. The edge case has no solution at all, since never vanishes.
Worked example 1.7.6
(Reading off the singularity at infinity)
For , has a pole of order at , matching the degree. For , has infinitely many negative powers, so infinity is essential, as the proposition predicts for a nonpolynomial. For a constant , is removable.
Worked example 1.7.7
(A cubic that is injective on but not on )
Let . On the real line , so is strictly increasing and injective there. On the theorem says it cannot be injective, since its degree is . Concretely, gives , so
The failure also appears locally: vanishes at , where is -to- near the point.
Practice and solutions
Quick check
(Classify the singularity of at zero and decide whether its image of a punctured disk is dense in .)
Solution · Answer
Its Laurent series has infinitely many negative powers, so zero is essential; Casorati–Weierstrass gives density on every punctured disk.
Quick check
(Can an entire injective function satisfy and ?)
Solution · Answer
No. Injectivity already forbids because ; the affine classification is not even needed.
Quick check
(If is entire and , what can be said about ?)
Solution · Answer
On , Cauchy's estimates give , which tends to as when . So every Taylor coefficient of degree greater than three vanishes, and is a polynomial of degree at most three.
Quick check
(Does the density theorem imply that takes the value zero near zero?)
Solution · Answer
No. The exponential is never zero; density only says every open neighborhood of zero meets the image.