A Taylor series is a local approximation: its disk of convergence can stop at a singularity even when the function is holomorphic far beyond that disk. Can one approximate on a compact set of a different shape? Cauchy's integral formula suggests replacing an integral by a finite sum of simple fractions. The location of the poles in those fractions then becomes a question about the holes in the compact set.
Throughout the approximation theorems, is nonempty and compact; write . The empty-set statements are vacuous. An approximation is uniform on ; it need not converge outside .
Definition 3.12.1
(Uniform approximation with poles in a prescribed set)
Let be compact and . A function on is uniformly approximable on by rational functions with poles in if for every there is a rational function whose finite poles all lie in and . For the approximants are exactly the polynomials.
From Cauchy integrals to rational functions
Theorem 3.12.2
(Rational approximation near a compact set)
Let be compact and let be holomorphic on an open neighborhood of . For every there is a rational function , with all its poles outside , such that .
Proof
(A grid and a uniform Riemann sum)
If is empty the result is immediate. Otherwise choose finitely many sufficiently small closed squares of one square grid: include every square that meets . Their union is compact, contains in its interior, and lies inside the given neighborhood. Choose the square diagonal smaller than , where is the given neighborhood; if , choose any positive diagonal. A point of in an edge belongs to both adjacent squares, and a point at a vertex belongs to all four, so each lies in the interior of and the corresponding common edges cancel.
Orient the boundary of each square counterclockwise and cancel common edges. The resulting finite polygonal cycle lies outside and has winding number one about each point of . Cauchy's formula for the cycle gives
The winding number is one on square interiors and is locally constant off , so the same holds on internal grid lines. Outside , the sum of the winding numbers of the square boundaries is zero; in particular it is zero outside , verifying the hypothesis of the cycle form of Cauchy's formula. Since and are disjoint compact sets, their distance is positive. The integrand is jointly continuous on , so sufficiently fine Riemann sums converge uniformly in . Each sum is a finite linear combination of with , hence a rational function with poles outside .
The existence of some exterior poles is not yet the strongest conclusion. We can move them without crossing .
Moving one pole through the complement
Theorem 3.12.3
(Runge's pole prescription)
Let be compact. Choose a set meeting every bounded connected component of . If is holomorphic on a neighborhood of , then it can be approximated uniformly on by rational functions whose finite poles all lie in .
In particular, if is connected, polynomials approximate uniformly on .
Proof
(Short steps, then a path)
Suppose and . The geometric expansion
converges uniformly for . Thus the function with a simple pole at is a uniform limit of rational functions with their only finite pole at . For a higher power , take the th power of approximants in (1); uniform convergence is preserved because uniformly convergent sequences of continuous functions on compact are uniformly bounded.
Now let and lie in the same component of the open set . Open connected subsets of the plane are path connected. Choose a continuous path from to ; its compact image has positive distance from . Subdivide the path so consecutive points are less than apart. Apply (1) successively. At each step a power of the previous simple pole is a uniform limit of polynomials in the next simple pole. Uniform limits of rational functions with poles in a fixed set are closed under sums, products and further uniform limits: the last property follows by choosing an approximant within of the th function. Induction along the finite chain therefore moves every power from to .
If lies in a bounded component, choose in that component. If lies in the unbounded component, connect it to a point with . There
uniformly on , so a pole at can be replaced by polynomials. The same is true for its powers. Partial fractions reduce any rational function with poles outside to finitely many such terms plus a polynomial. The number of terms is fixed and finite; choose each approximation error small enough that the sum is below the prescribed tolerance, then add them. Combining this with the first theorem proves the result. If the complement is connected, it is the unbounded component, so no finite poles are needed.
Proof X-ray
Why a path is enough
A single geometric expansion need not move a pole across an entire component: its ratio must be uniformly below one. A path stays a positive distance from the compact set, so finitely many short moves are permitted. Closure under uniform limits and products handles the finite chain; no infinite sequence of pole moves is needed.
The proof is constructive, and on a simple compact set the chain can be written down explicitly.
Worked example 3.12.4
(Two moves for a segment)
Let and . The complement of is connected, so polynomials must suffice. First move the pole from to : here and , so (1) applies with ratio , of modulus at most on :
Second, , so uniformly on with ratio at most ; powers of these partial sums approximate each . Given , choose with , then approximate the fixed terms to total error below . The result is a polynomial within of on .
Common mistake
(Runge polynomials are not Taylor partial sums)
In the segment example the Taylor series of about has radius , and at its terms have modulus , so it diverges at both endpoints of . No single power series can do better: one that converged at would converge on the open unit disk, agree with on , and hence be the Taylor series itself. The approximating polynomials change their coefficients as the tolerance shrinks.
What a hole prevents
Worked example 3.12.5
(A disk and a nearby exterior singularity)
Let and . Here the usual geometric expansion already provides the approximating polynomials
The bound comes from the geometric tail. For a function holomorphic near this closed disk, its Taylor series about zero also converges uniformly on the disk. Runge's theorem reaches more general compact shapes.
When has a hole, the prescription allows one pole in it, and the construction then produces a Laurent-type approximant.
Worked example 3.12.6
(One pole in the hole, one outside)
Let , and . The pole lies in the bounded component , which contains ; since , one step of (1) suffices:
The tail after is at most . The pole lies in the unbounded component with , so has error at most . Hence
and the only finite pole of is . This is the Laurent expansion of on , truncated in both directions.
Counterexample mode
A pole inside a hole cannot always be removed
Let and . The function is holomorphic on a neighborhood of , but no sequence of polynomials converges uniformly to it on . For any polynomial ,
Thus the error is at least . The bounded complementary component contains the pole , and allowing that pole makes the rational approximation exact. A hole permits an obstruction; it does not mean that every function on the compact set requires a pole there.
Common mistake
(A hole does not force a pole for each function)
The same annulus has a hole, but is already a polynomial. The theorem gives a sufficient prescription for all functions holomorphic near , not a claim that every allowed pole is necessary for each individual function.
More generally, every bounded complementary component is necessary in a prescription that works for all functions holomorphic near . Let be such a component, pick , and suppose no allowed pole lies in . The boundary lies in . Every rational approximant with allowed poles is holomorphic on a neighborhood of : it has no pole in or on its boundary. The maximum modulus principle applied to on bounded gives
The left side is the value at , so the error has a positive lower bound. Thus omitting prevents approximation of at least this one function.
Practice
Quick check
(On , which polynomials approximate , and how large is the error after degree ?)
Expand about and sum the geometric tail.
Solution · Answer
Take . On the tail is at most .
Quick check
(On , why does every polynomial satisfy ?)
Multiply by before integrating over the unit circle.
Solution · Answer
On the unit circle integrate . The polynomial term integrates to zero, while integrates to . Since there, the integral estimate gives .
Quick check
(If is finite, why can every function on be matched exactly by a polynomial?)
Only finitely many values are prescribed.
Solution · Answer
For distinct points with assigned values , Lagrange interpolation gives , and . An empty set is trivial. No neighborhood assumption is needed because only finitely many values are prescribed.
Exercise 3.12.7
(For , move a simple pole from to and bound the error of the sum through uniformly on .)
Core
Solution(3)
- 1
Check the short-step condition
Here and , so the ratio has modulus at most on .
Because: Expansion (1) needs .
- 2
Write the approximant
Key step
- 3
Bound the geometric tail
Since on ,
Exercise 3.12.8
(Suppose continuous functions and uniformly on compact . Prove uniformly on , and explain why this matters when moving higher-order poles.)
FoundationProof
Solution(3)
- 1
Split the error
- 2
Bound the factors
Key step
and are bounded on , and is bounded for large .
Because: Uniform limits of continuous functions on a compact set are continuous, hence bounded.
- 3
Conclude
∎Both terms tend to zero, so uniformly. Applied repeatedly, it shows that th powers of approximants of approximate , which is how higher-order poles are moved.