The integers are precisely the zeros of , all simple. A product with those zeros is easy to propose, but zeros alone leave an unknown zero-free entire factor. The logarithmic derivative detects that factor: a residue calculation for cotangent will identify it completely.
A partial fraction identity
Theorem 1.15.1
(Partial fractions of the cotangent)
For ,
The series converges locally uniformly on .
Pairing the positive and negative poles is essential: on with , , so each paired term is at most , and the Weierstrass -test applies to the tail. The identity itself is proved by residues.
Proof
(Residues on expanding squares)
Fix . Integrate
around the positively oriented square with sides and , where . Cotangent is uniformly bounded on these boundaries: on the vertical sides , of modulus at most , and on the horizontal sides . On we have , so , while the perimeter is . Hence the integral tends to zero.
At an integer , has residue , so the integrand has residue . At the residues are each, with sum . The residue theorem gives, for each such ,
Letting yields , and multiplying by gives (1).
Proof X-ray
Why squares through half-integers, and where z is used
The contour is chosen so that two estimates hold uniformly in . Passing through half-integers keeps the square at distance from every pole of , which is what makes bounded on independently of ; a contour through or near integers would lose this bound. The factor supplies decay of order , which beats the perimeter of order . With only the same crude estimate would give rather than , which is why the even denominator is used.
The hypothesis keeps the poles at distinct from the integer poles, so all residues are simple and computable. The pairing in (1) is not imposed afterwards: the residues at and arrive together inside each symmetric square, and that is why the identity comes out as a paired series.
Recovering the entire function
Integrating the logarithmic derivative now recovers sine from its zeros.
Theorem 1.15.2
(Product formula for the sine)
The product
converges locally uniformly on , and
where the left side is defined by its removable value at .
Proof
(Match logarithmic derivatives, then a constant)
Since , the product (2) converges locally uniformly, and its zeros are the integers, each simple. The partial products and their derivatives converge locally uniformly; away from the zeros, their logarithmic derivatives therefore converge to , and these logarithmic derivatives are the partial sums of the right side of (1). Hence off the integers. The quotient extends across the simple zeros to a zero-free entire function, and off the integers, so everywhere and is constant. Comparing derivatives at zero, and , so and . Dividing by gives (3).
Counterexample mode
Zeros do not determine the missing factor
The functions and have exactly the same zeros with the same multiplicities. A zero list alone cannot distinguish them. Their logarithmic derivatives differ by the constant , so equation (1) provides the extra information that makes the quotient constant; the limit of the quotient at zero then fixes that constant.
Common mistake
(Forgetting the normalizing constant)
The product equals , not and not . The function has derivative at , while the product has derivative ; and vanishes at the points , not at the integers. Check a candidate identity at before using it.
Worked examples
Worked example 1.15.3
(Reciprocal squares)
Near , the product in (3) has coefficient of equal to . This follows from locally uniform convergence: the -coefficient of a partial product is , and coefficients of locally uniformly convergent holomorphic functions converge. The Taylor series of the left side begins , so
The complex-analytic proof also controls the product on every compact set; it is not merely an identity at real points.
Worked example 1.15.4
(Wallis's product at )
At the left side of (3) is , and . Hence
The first three factors give about , compared with : the convergence is slow, since the th factor is .
Worked example 1.15.5
(A non-real point: )
Identity (1) holds at every , including . There , and . So (1) reads
and therefore .
Practice with solutions
Quick check
(Use (1) at to find .)
What is ?
Solution · Answer
Since and at , (1) gives , so the sum is .
Quick check
(Using (3), evaluate .)
Choose the value of .
Solution · Answer
Take : the product equals .
Quick check
(Why is the series of paired terms in (1) locally uniformly convergent away from integers?)
Bound one paired term on .
Solution · Answer
On a fixed compact set the tail is bounded by a constant multiple of ; the finitely many initial terms are continuous and bounded there. The Weierstrass -test applies to the tail.
Quick check
(What changes if the factors in (2) include an additional zero-free factor ?)
Compute the new logarithmic derivative.
Solution · Answer
The zeros are unchanged, but the logarithmic derivative gains and the function becomes .