Egorychev method
≈ 10 min read
Names in other languages
No names in other languages are held for this entry yet.
Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.
The Egorychev method is a collection of techniques introduced by Georgy Egorychev for finding identities among sums of binomial coefficients, Stirling numbers, Bernoulli numbers, Harmonic numbers, Catalan numbers and other combinatorial numbers. The method relies on two observations. First, many identities can be proved by extracting coefficients of generating functions. Second, many generating functions are convergent power series, and coefficient extraction can be done using the Cauchy residue theorem (usually this is done by integrating over a small circular contour enclosing the origin). The sought-for identity can now be found using manipulations of integrals. Some of these manipulations are not clear from the generating function perspective. For instance, the integrand is usually a rational function, and the sum of the residues of a rational function is zero, yielding a new expression for the original sum. The residue at infinity is particularly important in these considerations. Should a series appear during summation that is not finite the contours must be chosen such as to make the series converge.…
The Egorychev method is a collection of techniques introduced by Georgy Egorychev for finding identities among sums of binomial coefficients, Stirling numbers, Bernoulli numbers, Harmonic numbers, Catalan numbers and other combinatorial numbers. The method relies on two observations. First, many identities can be proved by extracting coefficients of generating functions. Second, many generating functions are convergent power series, and coefficient extraction can be done using the Cauchy residue theorem (usually this is done by integrating over a small circular contour enclosing the origin). The sought-for identity can now be found using manipulations of integrals. Some of these manipulations are not clear from the generating function perspective. For instance, the integrand is usually a rational function, and the sum of the residues of a rational function is zero, yielding a new expression for the original sum. The residue at infinity is particularly important in these considerations. Should a series appear during summation that is not finite the contours must be chosen such as to make the series converge. Some of the integrals employed by the Egorychev method are: First binomial coefficient integral {n\choose k} = \underset{z}{\mathrm{res}} \; \frac{(1+z)^n}{z^{k+1}} = \frac{1}{2\pi i} \int_{|z|=\rho} \frac{(1+z)^n}{z^{k+1}} \; dz where 0 Second binomial coefficient integral {n\choose k} = \underset{z}{\mathrm{res}} \; \frac{1}{(1-z)^{k+1} z^{n-k+1}} = \frac{1}{2\pi i} \int_{|z|=\rho} \frac{1}{(1-z)^{k+1} z^{n-k+1}} \; dz where 0 Exponentiation integral n^k = k! \; \underset{z}{\mathrm{res}} \; \frac{\exp(nz)}{z^{k+1}} = \frac{k!}{2\pi i} \int_{|z|=\rho} \frac{\exp(nz)}{z^{k+1}} \; dz: where 0 Iverson bracket k \le n = \underset{z}{\mathrm{res}} \; \frac{z^{k}}{z^{n+1}}\frac{1}{1-z} = \frac{1}{2\pi i} \int_{|z|=\rho} \frac{z^{k}}{z^{n+1}}\frac{1}{1-z} \; dz where 0 Stirling number of the first kind \left[ {n\atop k} \right] = \frac{n!}{k!} \; \underset{z}{\mathrm{res}} \; \frac{1}{z^{n+1}} \left(\log\frac{1}{1-z}\right)^k = \frac{n!}{k!} \frac{1}{2\pi i} \int_{|z|=\rho} \frac{1}{z^{n+1}} \left(\log\frac{1}{1-z}\right)^k \; dz where 0 Stirling number of the second kind \left\{ {n\atop k} \right\} = \frac{n!}{k!} \; \underset{z}{\mathrm{res}} \; \frac{(\exp(z)-1)^k}{z^{n+1}} = \frac{n!}{k!} \frac{1}{2\pi i} \int_{|z|=\rho} \frac{(\exp(z)-1)^k}{z^{n+1}} \; dz where 0 S_j(n) = \sum_{k=0}^n (-1)^k {n\choose k} {n+k\choose k} {k\choose j} which is claimed to be : (-1)^n {n\choose j}{n+j\choose j}. Introduce : {n+k\choose k} = \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^{n+k}}{z^{k+1}} \; dz and : {k\choose j} = \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{(1+w)^k}{w^{j+1}} \; dw. This yields for the sum : \begin{align} & \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^n}{z} \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^{j+1}} \sum_{k=0}^n (-1)^k {n\choose k} \frac{(1+z)^k (1+w)^k}{z^k} \; dw \; dz \\[6pt] = {} & \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^{n}}{z} \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^{j+1}} \left(1-\frac{(1+w)(1+z)}{z}\right)^n \; dw \; dz \\[6pt] = {} & \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^{n}}{z^{n+1}} \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^{j+1}} (-1-w-wz)^n \; dw \; dz \\[6pt] = {} & \frac{(-1)^n}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^n}{z^{n+1}} \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^{j+1}} (1+w+wz)^n \; dw \; dz. \end{align} This is \frac{(-1)^n}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^{n}}{z^{n+1}} \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^{j+1}} \sum_{q=0}^n {n\choose q} w^q (1+z)^q \; dw \; dz. Extracting the residue at w=0 we get \begin{align} & \frac{(-1)^n}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^{n}}{z^{n+1}} {n\choose j} (1+z)^j \; dz \\[6pt] = {} & {n\choose j} \frac{(-1)^n}{2\pi i} \int_{|z|=\varepsilon} \frac{(1+z)^{n+j}}{z^{n+1}}\; dz \\[6pt] = {} & (-1)^n {n\choose j} {n+j\choose n} \end{align} thus proving the claim. There are no convergence issues here as the sums involved are finite and with n+k and k not being negative we can choose any non-zero finite value for \varepsilon and \gamma . Suppose we seek to evaluate \sum_{k=1}^n k {2n\choose n+k}. Introduce {2n\choose n+k} = \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{1}{z^{n-k+1}} \frac{1}{(1-z)^{n+k+1}} \; dz. Observe that this is zero when k> n so we may extend k to infinity to obtain for the sum \begin{align} & \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{1}{z^{n+1}} \frac{1}{(1-z)^{n+1}} \sum_{k\ge 1} k \frac{z^k}{(1-z)^k} \; dz \\[6pt] = {} & \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{1}{z^{n+1}} \frac{1}{(1-z)^{n+1}} \frac{z/(1-z)}{(1-z/(1-z))^2} \; dz \\[6pt] = {} & \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{1}{z^{n}} \frac{1}{(1-z)^n} \frac{1}{(1-2z)^2} \; dz. \end{align} Now put z(1-z)=w so that (observe that with w=z+\cdots the image of |z|=\varepsilon with \varepsilon small is another closed circle-like contour which makes one turn and which we may certainly deform to obtain another circle |w|=\gamma ) z = \frac{1-\sqrt{1-4w}}{2} \quad\text{and}\quad (1-2z)^2 = 1-4w and furthermore dz = -\frac{1}{2} \times \frac{1}{2} \times (-4) \times (1-4w)^{-1/2} \; dw = (1-4w)^{-1/2} \; dw to get for the integral \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^n} \frac{1}{1-4w} (1-4w)^{-1/2} \; dw = \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{1}{w^n} \frac{1}{(1-4w)^{3/2}} \; dw. This evaluates by inspection to (use the Newton binomial) \begin{align} & 4^{n-1} {n-1+1/2\choose n-1} = 4^{n-1} {n-1/2\choose n-1} = \frac{4^{n-1}}{(n-1)!} \prod_{q=0}^{n-2} (n-1/2-q) \\ = {} & \frac{2^{n-1}}{(n-1)!} \prod_{q=0}^{n-2} (2n-2q-1) = \frac{2^{n-1}}{(n-1)!} \frac{(2n-1)!}{2^{n-1} (n-1)!} \\[6pt] = {} & \frac{n^2}{2n} {2n\choose n} = \frac{1}{2} n {2n\choose n}. \end{align} Here the mapping from z=0 to w=0 determines the choice of square root. For the conditions on \epsilon and \gamma we have that for the series to converge we require |z/(1-z)| or \epsilon/(1-\epsilon) or \epsilon The closest that the image contour of |z|=\epsilon comes to the origin is \epsilon-\epsilon^2 so we choose \gamma for example \gamma = \epsilon^2-\epsilon^3. This also ensures that \gamma so |w|=\gamma does not intersect the branch cut [1/4,\infty) (and is contained in the image of |z|=\epsilon ). For example \epsilon = 1/3 and \gamma = 2/27 will work. This example also yields to simpler methods but was included here to demonstrate the effect of substituting into the variable of integration. the poles at z=1 and z=1/2 are not inside the contour since \varepsilon ): \frac{1}{2\pi i} \int_{|z|=\varepsilon} \frac{1}{z^{n}} \frac{1}{(1-z)^n} \frac{1}{(1-2z)^2} \; dz = \underset{z}{\mathrm{res}} \frac{1}{z^{n}} \frac{1}{(1-z)^n} \frac{1}{(1-2z)^2} with A(z) = \frac{z}{(1-2z)^2} and f(z) = \frac{1}{1-z}. We get h(z) = z (1-z) and find \underset{w}{\mathrm{res}} \frac{1}{w^{n+1}} \left.\left[ \frac{A(z)}{f(z) h'(z)} \right]\right|_{z=g(w).} with g the inverse of h . This becomes \underset{w}{\mathrm{res}} \frac{1}{w^{n+1}} \left.\left[ \frac{z/(1-2z)^2}{(1-2z)/(1-z)} \right]\right|_{z=g(w)} or alternatively \underset{w}{\mathrm{res}} \frac{1}{w^{n+1}} \left.\left[ \frac{z(1-z)}{(1-2z)^3} \right]\right|_{z=g(w)} = \underset{w}{\mathrm{res}} \frac{1}{w^{n}} \left.\left[ \frac{1}{(1-2z)^3} \right]\right|_{z=g(w).} Observe that (1-2z)^2 = 1 - 4z + 4z^2 = 1-4z(1-z) = 1-4w so this is \underset{w}{\mathrm{res}} \frac{1}{w^{n}} \frac{1}{(1-4w)^{3/2}} and the rest of the computation continues as before.
Text derived from https://en.wikipedia.org/w/index.php?oldid=1370489964 under CC BY-SA 4.0; modifications have been made.