Jacobi Triple Product Identity

Jacobi’s triple product is one of the basic identities in the theory of theta functions and $q$ -series. It converts a bilateral theta series, indexed by all integers, into an infinite product. In the form we shall use, it say:

For ${|q|<1}$ and ${x \neq 0}$ , we have

$\displaystyle \sum_{n=-\infty}^{\infty} q^{n^{2}} x^{n}=\prod_{n=1}^{\infty}\left(1+x q^{2 n-1}\right)\left(1+x^{-1} q^{2 n-1}\right)\left(1-q^{2 n}\right)$

The identity is powerful because the two sides look as though they belong to different worlds. The left side is additive: it is a sum over all integer powers of $x$ , weighted by the quadratic exponent $n^2$ . The right side is multiplicative: it is an infinite product whose factors visibly encode zeros at a regular geometric sequence of points. The proof below explains why these two expressions are forced to be the same. The series side has functional equations that determine its zeros; the product side is precisely the product with those zeros; after dividing them, nothing nonconstant is left.

Define

$F(x) =\sum_{n=-\infty}^{\infty} q^{n^{2}} x^{n}.$

For $|q|<1$ , this series converges absolutely and uniformly on compact annuli $0<c\le |x|\le C<\infty$ , so it defines a holomorphic function of $x$ on the punctured plane $\mathbf C^\times$ . The function has two elementary symmetries. First, replacing $n$ by $-n$ , that is switching the terms ${a_n}$ and ${a_{-n},}$ , we get

$\displaystyle F(x) = F\left(\frac{1}{x}\right )$

Second, shifting the index gives a multiplicative functional equation. Indeed,

$\displaystyle F(x) =\sum_{n=-\infty}^{\infty} q^{(n+1)^{2}} x^{n+1}=qx\sum_{n=-\infty}^{\infty}q^{n^2}(q^2x)^n = q x F\left(q^{2} x\right)$

These two equations already reveal the zeros. Put $x=-q^{-1}$ . The symmetry gives

$\displaystyle F(-q^{-1})=F(-q).$

The functional equation gives

$q(-q^{-1})F(-q) =-F(-q).$

Therefore $F(-q)=-F(-q)$ , and hence

$\displaystyle F(-q) =F(-\frac{1}{q})=0.$

Applying the same functional equations repeatedly propagates this zero along the full two-sided geometric progression. In fact apply the functional relations again to get,

$\displaystyle F(-q^{2n-1})=0, \quad \mbox{for all}~~ n \in \mathbb{Z}$

Equivalently, the zeros occur at $-q, -q^3, -q^5,\ldots$ and also at their reciprocals $-q^{-1}, -q^{-3}, -q^{-5},\ldots.$

This is the conceptual heart of the product. If a function on $\mathbf C^\times$ has these zeros, the natural infinite product carrying exactly those zeros is

$\displaystyle P(x)= \prod_{n=1}^{\infty} (1+xq^{2n-1})(1+x^{-1}q^{2n-1}).$

The first factor vanishes when $x=-q^{-(2n-1)}$ , while the second vanishes when $x=-q^{2n-1}$ . Thus the two halves of the product encode the two halves of the zero set. Since the denominator has only simple zeros and $F$ vanishes at all of them, the quotient

$\displaystyle G(x)=\dfrac{F(x)}{P(x)} =\dfrac{F(x)}{\displaystyle \prod_{n=1}^{\infty}\left(1+x q^{2 n-1}\right)\left(1+x^{-1} q^{2 n-1}\right)}$

has no poles on $\mathbf C^\times$ .

Now we use the functional equation to show that this quotient is constant. First compute how $P$ changes under $x\mapsto q^2x$ . One has

$\displaystyle P(q^2 x)= \prod_{n=1}^{\infty} (1+xq^{2n+1})(1+x^{-1}q^{2n-3}).$

The first product is the original positive product with the first factor $1+xq$ removed. The second product is the original reciprocal product with one extra factor $1+x^{-1}q^{-1}$ inserted. Hence

$\displaystyle P(q^2x)= P(x)\frac{1+x^{-1}q^{-1}}{1+xq} = \frac{P(x)}{qx}.$

Since $F(q^2x)=F(x)/(qx)$ , we obtain

$\displaystyle G(q^2x)= G(x).$

Thus $G$ is holomorphic on $\mathbf C^\times$ and invariant under multiplication of the argument by $q^2$ . Write its Laurent expansion about $0$ :

$\displaystyle G(x) =\sum_{k=0}^{\infty} A_k(q) x^n.$

The equation $G(q^2x)=G(x)$ gives

$\displaystyle (1-q^2k)A_{k}(q)=0$

Since $|q|<1$ , we have $q^{2k}\ne1$ for every $k\ne0$ . Hence all nonzero Laurent coefficients vanish, and

$\displaystyle G(x) = A_0(q).$

Thus we have proved that

$\displaystyle F(x) = A_0(q)P(x)= A_0(q)\prod_{n=1}^{\infty}\left(1+x q^{2 n-1}\right)\left(1+x^{-1} q^{2 n-1}\right).$

Everything now depends on finding the scalar factor $A_0(q)$ . This is the only missing piece. The product has the correct zeros and the correct functional equation; the remaining ambiguity is a factor depending on $q$ alone.

To determine it, compare two specializations of the same theta series. First observe that

$\displaystyle F_q(i)=\sum_{n=-\infty}^{\infty}q^{n^2}i^n.$

The odd terms cancel in pairs, because $i^n+i^{-n}=0$ when $n$ is odd. The even terms have $n=2m$ , and then $i^{2m}=(-1)^m$ . Hence

$\displaystyle F_q(i) =\sum_{m=-\infty}^{\infty} q^{4 m^{2}}(-1)^{m}$

Now use the product representation at $x=i$ , to get

$\displaystyle \sum_{n=-\infty}^{\infty} q^{4 n^{2}}(-1)^{n}=A_{0}(q) \prod_{n=1}^{\infty}\left(1+q^{4 n-2}\right)$

At the same time, using the parameter $q^4$ and setting $x=-1$ gives

$\displaystyle \sum_{n=-\infty}^{\infty} q^{4 n^{2}}(-1)^{n}=A_{0}\left(q^{4}\right) \prod_{n=1}^{\infty}\left(1-q^{8 n-4}\right)^{2}$

Thus we have

$\displaystyle A_{0}(q) \prod_{n=1}^{\infty}\left(1+q^{4 n-2}\right) = A_{0}\left(q^{4}\right) \prod_{n=1}^{\infty}\left(1-q^{8 n-4}\right)^{2}$

Using $1-q^{8n-4}=(1-q^{4n-2})(1+q^{4n-2}),$ this becomes

$\displaystyle A_{0}(q) =A_0(q^4) \prod_{n=1}^{\infty}(1-q^{4n-2})(1-q^{8n-4}).$

The product on the right is exactly the quotient of the even Euler products

$\displaystyle \dfrac{ \displaystyle \prod_{n=1}^{\infty}(1-q^{2n})} { \displaystyle \prod_{n=1}^{\infty}(1-q^{8n})}.$

Therefore, we have

$\displaystyle A_{0}(q) =A_0(q^4)~ \dfrac{ \displaystyle \prod_{n=1}^{\infty}(1-q^{2n})} { \displaystyle \prod_{n=1}^{\infty}(1-q^{8n})}.$

and hence

$\displaystyle \dfrac{A_{0}(q)}{\displaystyle \prod_{n=1}^{\infty}\left(1-q^{2 n}\right)}=\dfrac{A_{0}\left(q^{4}\right)}{\displaystyle \prod_{n=1}^{\infty}\left(1-q^{8 n}\right)} .$

Iterate this map ${q \rightarrow q^4},$ we see that

$\displaystyle \dfrac{A_{0}(q)}{\displaystyle \prod_{n=1}^{\infty}\left(1-q^{2 n}\right)}=\dfrac{A_{0}\left(q^{4^k}\right)}{\displaystyle \prod_{n=1}^{\infty}\left(1-q^{2.4^k n}\right)}\longrightarrow_{k \rightarrow \infty} 1$

Since $|q|<1$ , we have $q^{4^m}\to0$ . As $q\to0$ , the series $F(x)$ tends to $1$ and the product $P(x)$ tends to $1$ , so $A_0(q)\to1$ . The denominator on the right also tends to $1$ . Hence the limiting value of the right-hand side is $1$ , and so

$\displaystyle A_0(q) =\displaystyle \prod_{n=1}^{\infty}\left(1-q^{2 n}\right)$

Substituting this back gives Jacobi’s triple product:

$\displaystyle \sum_{n=-\infty}^{\infty} q^{n^{2}} x^{n}=\prod_{n=1}^{\infty}\left(1+x q^{2 n-1}\right)\left(1+x^{-1} q^{2 n-1}\right)\left(1-q^{2 n}\right)$

The proof should be read as an analogue of the product formula for sine. There, the zeros of $\sin \pi z$ force a product over the integers. Here, the zeros of a theta function on the multiplicative plane force a product over the geometric progression $-q^{2m+1}$ . The extra factor $\prod(1-q^{2n})$ is the normalization constant.

Euler’s one-sided product expansions

There is another very useful way to prove Jacobi’s triple product. The first proof used the zeros and functional equations of the theta series. This second proof is more algebraic. It begins with two one-sided Euler product expansions and then shows that, when these expansions are combined in the right way, the two-sided theta series appears automatically. These products are simpler than the full triple product because they involve only nonnegative powers of $x$ . The remarkable point is that, after multiplying the right one-sided expansions together, the negative powers of $x$ appear automatically, and the bilateral theta series emerges.

Consider the product

$\displaystyle A(q,x)=\prod_{n=1}^{\infty}(1+xq^{2n-1}).$

This product satisfies a simple functional equation. The first factor is $1+qx$ , and after removing it, the remaining product is obtained by replacing $x$ with $q^2x.$ Therefore

$\displaystyle A(q,x)=(1+qx)A(q,q^2x),$

Now expand $A(q,x)$ as a power series in $x$ :

$\displaystyle A(q,x)=\sum_{n=0}^{\infty}c_nx^n.$

Substituting this power series into the functional equation gives

$\displaystyle A(q,x)=(1+qx)\sum_{n=0}^{\infty}c_nq^{2n}x^n.$

The coefficient of $x^n$ on the right receives one contribution from the first part and one contribution from multiplying by $qx$ . Thus, for $n\ge1$ ,

$\displaystyle c_n=q^{2n}c_n+q^{2n-1}c_{n-1}.$

Hence $c_n(1-q^{2n})=q^{2n-1}c_{n-1}.$ Since $c_0=1$ , iteration yields

$\displaystyle c_n= \dfrac{q^{n^2}}{\prod_{k=1}^{n}(1-q^{2k})}.$

Thus Euler’s first expansion is

$\displaystyle A(q,x)= \prod_{n=1}^{\infty}(1+xq^{2n-1})=\sum_{n=0}^{\infty} \frac{q^{n^2}x^n}{\prod_{k=1}^{n}(1-q^{2k})}.$

Next consider the reciprocal product

$\displaystyle B(q,x)=\prod_{n=1}^{\infty}(1+xq^{2n-1})^{-1}.$

It satisfies the functional equation

$\displaystyle B(q,x)=(1+qx)^{-1}B(q,q^2x).$

Equivalently,

$\displaystyle (1+qx)B(q,x)=B(q,q^2x).$

Write the power series expansion

$\displaystyle B(q,x)=\sum_{n=0}^{\infty}d_nx^n.$

Then

$\displaystyle B(q,q^2x) =\sum_{n=0}^{\infty}d_nq^{2n}x^n.$

Comparing coefficients of $x^n$ , for $n\ge1$ , gives

$\displaystyle d_n+qd_{n-1}=q^{2n}d_n.$

Therefore

$\displaystyle d_n(1-q^{2n})=-qd_{n-1}.$

Since $d_0=1$ , we obtain

$\displaystyle d_n= \dfrac{(-1)^nq^n}{\prod_{k=1}^{n}(1-q^{2k})}.$

Thus Euler’s second expansion is

$\displaystyle B(q,x)= \prod_{n=1}^{\infty}(1+xq^{2n-1})^{-1}=\sum_{n=0}^{\infty} \frac{(-1)^nq^nx^n}{\prod_{k=1}^{n}(1-q^{2k})}.$

Now we use the two Euler expansions to reconstruct the full triple product. The basic point is that $A(q,x)$ only contains nonnegative powers of $x$ , while the theta series

$\displaystyle \Theta(q,x)=\sum_{m=-\infty}^{\infty}q^{m^2}x^m$

contains all integer powers of $x$ .

The reciprocal product $B(q,x^{-1})$ will supply the negative powers, and the Euler factor

$\displaystyle E(q)=\prod_{n=1}^{\infty}(1-q^{2n}).$

will clear the denominators which appeared in Euler’s expansion of $A(q,x)$ .

Multiplying the first Euler expansion by $E(q)$ clears the finite denominator in each coefficient. We get

$\displaystyle A(q, x) E(q)=\sum_{n=0}^{\infty} q^{n^2}x^n \prod_{j=n+1}^{\infty}(1-q^{2j}).$

The tail product can be rewritten as another value of $A$ . Indeed, it is equal to

$\displaystyle \prod_{k=0}^{\infty}(1-q^{2n+2k+2}),$

which is exactly $A(q,-q^{2n+1}) .$ To see this the Euler’s substitute $x=-q^{2n+1}$ in the product expansion of $A(q, x)$ , which gives

$\displaystyle \prod_{k=1}^{\infty}(1-q^{2n+1}q^{2k-1}) =\prod_{k=1}^{\infty}(1-q^{2n+2k})=\prod_{k=0}^{\infty}(1-q^{2n+2k+2}),.$

Using Euler’s first expansion again, with $x=-q^{2n+1}$ , gives

$\displaystyle A(q,-q^{2n+1})= \sum_{k=0}^{\infty} \frac{(-1)^kq^{k^2+(2n+1)k}}{\prod_{\ell=1}^{k}(1-q^{2\ell})}.$

Substituting this into the expression for $E(q)A(q,x)$ gives

$\displaystyle E(q)A(q,x)=\sum_{n=0}^{\infty}\sum_{k=0}^{\infty} \frac{(-1)^kq^{n^2+k^2+(2n+1)k}x^n}{\prod_{\ell=1}^{k}(1-q^{2\ell})} = \sum_{n=0}^{\infty}\sum_{k=0}^{\infty} \frac{(-1)^kq^{(n+k)^2+k}x^n}{\prod_{\ell=1}^{k}(1-q^{2\ell})}.$

We now compare this with the product $\Theta(q,x)B(q,x^{-1})$ . The factor

$\displaystyle \frac{(-1)^kq^kx^{-k}}{\prod_{\ell=1}^{k}(1-q^{2\ell})}$

is the $k$ th term in the expansion of $B(q,x^{-1})$ . Therefore the expression above is precisely the nonnegative-power part of the product

$\displaystyle \left(\sum_{m=-\infty}^{\infty}q^{m^2}x^m\right)B(q,x^{-1}).$

Look first at the coefficient of a nonnegative power $x^n$ , where $n\ge0$ . The condition $m-k=n$ means $m=n+k$ . Hence the coefficient of $x^n$ in $\Theta(q,x)B(q,x^{-1})$ is

$\displaystyle \sum_{k=0}^{\infty} \frac{(-1)^kq^{(n+k)^2+k}} {\prod_{\ell=1}^{k}(1-q^{2\ell})},$

which is exactly the coefficient of $x^n$ in the expansion of $E(q)A(q,x)$ above. It remains to check that $\Theta(q,x)B(q,x^{-1})$ has no negative powers of $x$ . Let $s>0$ . The coefficient of $x^{-s}$ is obtained from the condition $m-k=-s$ , so $m=k-s$ . Therefore that coefficient is

$\displaystyle \sum_{k=0}^{\infty} \frac{(-1)^kq^{(k-s)^2+k}} {\prod_{\ell=1}^{k}(1-q^{2\ell})}.$

Thus the coefficient is

$\displaystyle q^{s^2} \sum_{k=0}^{\infty} \frac{(-1)^kq^{k^2+(1-2s)k}} {\prod_{\ell=1}^{k}(1-q^{2\ell})}.$

But the sum is exactly Euler’s expansion for $A(q,-q^{1-2s})$ . Hence the coefficient of $x^{-s}$ is

$\displaystyle q^{s^2}A(q,-q^{1-2s}).$

But $A(q,-q^{1-2s})$ contains the factor

$\displaystyle 1-q^{1-2s}q^{2s-1}=0,$

so every negative-power coefficient vanishes. Hence we proved

$\displaystyle \Theta(q,x) B(q,x^{-1}) = A(q, x) E(q).$

Now multiply both sides by $A(q,x^{-1})$ . Since $B(q,x^{-1})$ is the reciprocal of $A(q,x^{-1})$ , the left-hand side becomes just $\Theta(q,x)$ :

$\displaystyle \sum_{m=-\infty}^{\infty}q^{m^2}x^m.$

Finally unfold the definitions. Since

$\displaystyle E(q)=\prod_{n=1}^{\infty}(1-q^{2n}),$

$\displaystyle A(q,x)=\prod_{n=1}^{\infty}(1+xq^{2n-1}),$

$\displaystyle A(q,x^{-1})=\prod_{n=1}^{\infty}(1+x^{-1}q^{2n-1}),$

we obtain that the right-hand side equals

$\displaystyle \prod_{n=1}^{\infty} (1+xq^{2n-1})(1+x^{-1}q^{2n-1})(1-q^{2n}).$

Thus we have

$\displaystyle \sum_{m=-\infty}^{\infty}q^{m^2}x^m =\prod_{n=1}^{\infty} (1+xq^{2n-1})(1+x^{-1}q^{2n-1})(1-q^{2n}).$

This is Jacobi’s triple product. The proof shows that the theta series is assembled from two one-sided Euler expansions: $A(q,x)$ produces the positive powers, $B(q,x^{-1})$ introduces the possible negative powers, and the apparent extra negative terms vanish because $A(q,-q^{1-2s})$ contains an explicit zero factor. That cancellation is the algebraic version of the zero-set argument in the first proof.

Euler’s pentagonal theorem

One of the most important consequences of Jacobi’s triple product is Euler’s pentagonal theorem. We obtain it by the substitution

$\displaystyle q\mapsto q^{3/2}, \quad x=-q^{1/2}.$

On the series side of the triple product, this gives

$\displaystyle \sum_{n=-\infty}^{\infty} (-1)^nq^{(3n^2+n)/2}.$

On the product side, the three factors become

$\displaystyle 1+x(q^{3/2})^{2n-1}=1-q^{3n-1}, \\\\ 1+x^{-1}(q^{3/2})^{2n-1}=1-q^{3n-2}, \\\\ 1-(q^{3/2})^{2n}=1-q^{3n}.$

Therefore the product side is

$\displaystyle \prod_{n=1}^{\infty} (1-q^{3n-1})(1-q^{3n-2})(1-q^{3n}).$

The exponents $3n-2$ , $3n-1$ , and $3n$ run through all positive integers, separated according to their residues modulo $3$ . Thus we get

$\displaystyle \prod_{n=1}^{\infty}(1-q^n).$

Hence

$\displaystyle \sum_{n=-\infty}^{\infty} (-1)^nq^{(3n^2+n)/2}= \prod_{n=1}^{\infty}(1-q^n)..$

Separating the $n=0$ term and then pairing the terms with indices $n$ and $-n$ gives the usual form

$\displaystyle 1+\sum_{n=1}^{\infty} (-1)^n \bigl( q^{n(3n-1)/2} + q^{n(3n+1)/2} \bigr)= \prod_{n=1}^{\infty}(1-q^n).$

This is Euler’s pentagonal theorem. The name comes from the exponents $\frac{n(3n-1)}{2}, \frac{n(3n+1)}{2},$ which are the generalized pentagonal numbers. From the triple-product viewpoint, these numbers are not mysterious. They arise from the quadratic exponent $n^2$ in the theta series after the substitution $q\mapsto q^{3/2}$ and $x=-q^{1/2}$ .

Euler also had a direct proof of the pentagonal theorem which does not pass through Jacobi’s triple product. The proof is based on a cleverly chosen auxiliary function. The purpose of this function is twofold. First, when we put $x=1$ , it becomes Euler’s infinite product $\prod_{n=1}^{\infty}(1-q^n)$ . Second, as a function of $x$ , it satisfies a simple functional equation. Iterating that functional equation produces exactly the two families of generalized pentagonal exponents.

Define

$\displaystyle f(q,x) =1-\sum_{n=1}^{\infty} (1-xq)(1-xq^2)\cdots(1-xq^{n-1})x^{n+1}q^n.$

For $n=1$ , the product is empty and is taken to be $1$ . The point of this definition is that, when $x=1$ , the series telescopes. Put

$\displaystyle P_n=\prod_{j=1}^{n}(1-q^j), \quad P_0=1.$

Then

$\displaystyle P_{n-1}q^n=P_{n-1}-P_n.$

Therefore

$\displaystyle f(q,1) =1-\sum_{n=1}^{\infty}(P_{n-1}-P_n) =\prod_{n=1}^{\infty}(1-q^n).$

So the function $f(q,x)$ really is a deformation of Euler’s product: at $x=1$ , it gives exactly the product whose series expansion we want.

The second key fact is that $f(q,x)$ satisfies the functional equation

$\displaystyle f(q,x)=1-x^2q-x^3q^2f(q,qx).$

This relation is obtained by separating the first term $x^2q$ from the defining series and then rewriting the remaining tail in terms of the same function with $x$ replaced by $qx$ .

Now we iterate the functional equation. The first step gives

$\displaystyle f(q,x)=1-x^2q-x^3q^2f(q,qx).$

Apply the same formula to $f(q,qx)$ :

$\displaystyle f(q,qx)=1-x^2q^3-x^3q^5f(q,q^2x).$

Substituting this into the previous line gives

$\displaystyle f(q,qx)= 1-x^2q-x^3q^2 +x^5q^5 +x^6q^7f(q,q^2x).$

Doing it once more gives the next two terms:

$\displaystyle f(q,qx)= 1-x^2q-x^3q^2 +x^5q^5+x^6q^7 -x^8q^{12} -x^9q^{15}f(q,q^3x).$

We can already see the pattern. Thus repeated substitution gives

$\displaystyle 1+\sum_{n=1}^{\infty} (-1)^n \bigl( x^{3n-1}q^{n(3n-1)/2} + x^{3n}q^{n(3n+1)/2} \bigr).$

Finally set $x=1$ . From the telescoping argument above, $f(q,1)=\prod_{n=1}^{\infty}(1-q^n)$ , so we get

$\displaystyle 1+\sum_{n=1}^{\infty} (-1)^n \bigl( x^{3n-1}q^{n(3n-1)/2} + x^{3n}q^{n(3n+1)/2} \bigr)= \prod_{n=1}^{\infty}(1-q^n) .$

This is Euler’s pentagonal theorem. The generalized pentagonal exponents appear because each iteration of the functional equation advances the powers of $x$ by blocks of three while accumulating the quadratic powers $n(3n-1)/2$ and $n(3n+1)/2$ .

Ramanujan’s general theta function

Ramanujan often wrote Jacobi’s triple product in a more symmetric two-parameter form. This notation is extremely useful because it treats the positive and negative directions of the bilateral sum on equal footing. Define

$\displaystyle f(a,b) =\sum_{n=-\infty}^{\infty} a^{n(n+1)/2}b^{n(n-1)/2}, \qquad |ab|<1.$

At first sight, this looks different from the theta series

$\displaystyle \sum_{n=-\infty}^{\infty}q^{n^2}x^n.$

But it is exactly the same object after a change of variables. Indeed, $a^{n(n+1)/2}b^{n(n-1)/2} = (ab)^{n^2/2}(a/b)^{n/2}.$ Thus, if we put $\displaystyle q=(ab)^{1/2}, ~ x=(a/b)^{1/2},$ then

$\displaystyle q^{n^2}x^n =(ab)^{n^2/2}(a/b)^{n/2}= a^{n(n+1)/2}b^{n(n-1)/2}.$

So Ramanujan’s function $f(a,b)$ is Jacobi’s theta series written in variables adapted to the two directions of summation.

Applying Jacobi’s triple product gives

$\displaystyle \sum_{n=-\infty}^{\infty} a^{n(n+1)/2}b^{n(n-1)/2} = \prod_{j=0}^{\infty}(1+a(ab)^j) \prod_{j=0}^{\infty}(1+b(ab)^j) \prod_{j=1}^{\infty}(1-(ab)^j).$

Using the $q$ -Pochhammer symbol

$\displaystyle (c;q)_\infty = \prod_{j=0}^{\infty}(1-cq^j).$

Ramanujan’s form of the triple product is

$\displaystyle f(a,b) = (-a;ab)_\infty(-b;ab)_\infty(ab;ab)_\infty.$

This form is often more flexible than the original $q,x$ version. The parameters $a$ and $b$ separate the two directions of the bilateral sum. Positive and negative shifts of the suation index naturally interchange the roles of $a$ and $b$ . This is why Ramanujan’s notation appears so often in theta-function identities: many important identities become simple substitutions for $a$ and $b$ . For example, Euler’s pentagonal theorem comes from choosing parameters so that the three product families collapse into the single Euler product $\prod_{n=1}^{\infty}(1-q^n)$ . Other choices lead to many classical theta identities and product-series transformations. Ramanujan’s $f(a,b)$ is therefore not a separate theorem from Jacobi’s triple product. It is the same theorem written in variables that make later specializations easier to recognize.

Watson’s quintuple product identity

Jacobi’s triple product is the basic product-sum identity for theta functions, but it is not the end of the story. A deeper identity in the same family is Watson’s quintuple product identity. One common form is

$\displaystyle \prod_{n=1}^{\infty} (1-q^n)(1-zq^n)(1-z^{-1}q^{n-1}) (1-z^2q^{2n-1})(1-z^{-2}q^{2n-1}) \\ = \sum_{m=-\infty}^{\infty} (z^{3m}-z^{-3m-1})q^{m(3m+1)/2}.$

This identity should be read as a more elaborate cousin of the triple product. In Jacobi’s identity, the product side has three structured families of factors: $1-q^{2n}, \quad 1+xq^{2n-1}, 1+x^{-1}q^{2n-1}.$ In Watson’s identity, the product side has five families of factors:

$\displaystyle 1-q^n, \quad 1-zq^n, \quad 1-z^{-1}q^{n-1}, \quad 1-z^2q^{2n-1}, \quad 1-z^{-2}q^{2n-1}.$

Thus the left side contains a richer zero pattern. The factors involving $z$ and $z^{-1}$ encode one pair of reciprocal geometric progressions, while the factors involving $z^2$ and $z^{-2}$ encode another pair, now along the odd powers of $q$ . The surprising fact is that all of this multiplicative structure collapses into a single bilateral series with quadratic exponent $\frac{m(3m+1)}{2}$ and coefficient $z^{3m}-z^{-3m-1}.$ This coefficient is important. In Jacobi’s triple product, each term of the theta series has a single monomial coefficient, namely $x^n$ . In Watson’s identity, the coefficient is a difference of two monomials. That difference reflects an additional cancellation built into the product. So the quintuple product is not just the triple product with more factors. It is a more refined identity in which two theta-like contributions are combined and many terms cancel, leaving the compact expression on the right. The same broad principle remains: a carefully structured infinite product can be unfolded into a bilateral series with quadratic exponents. The triple product is the fundamental case. The quintuple product is a richer case, where the zero structure is more complicated and the series side records this complication through the coefficient $z^{3m}-z^{-3m-1}$ .

Jacobi’s triple product therefore lies at a crossroads of several subjects. Analytically, it is a factorization theorem for theta functions. Algebraically, it is an identity between an infinite product and a bilateral series. Combinatorially, it generates partition identities such as Euler’s pentagonal-number theorem. In the theory of modular forms, it is one of the fundamental bridges between theta series and infinite products. Its importance comes from this translating power. The triple product converts additive information, expressed through a sum over integers, into multiplicative information, expressed through an infinite product. It is not merely one remarkable identity among many. It is a general mechanism for passing between the additive and multiplicative structures that govern the theory of $q$ -series.

Share this:

Related

Leave a comment Cancel reply