Vol. 37 (2006) ACTA PHYSICA POLONICA B No 10 EVALUATING RESIDUES AND INTEGRALS THROUGH NEGATIVE DIMENSIONAL INTEGRATION METHOD (NDIM) Alfredo Takashi Suzuki Instituto de Física Teórica, Universidade Estadual Paulista Rua Pamplona, 145-01405-900 São Paulo, SP, Brasil (Received May 4, 2006 ) The standard way of evaluating residues and some real integrals through the residue theorem (Cauchy’s theorem) is well-known and widely applied in many branches of Physics. Herein we present an alternative technique based on the negative dimensional integration method (NDIM) originally developed to handle Feynman integrals. The advantage of this new tech- nique is that we need only to apply Gaussian integration and solve systems of linear algebraic equations, with no need to determine the poles them- selves or their residues, as well as obtaining a whole class of results for differing orders of poles simultaneously. PACS numbers: 01.55.+b, 02.30.Cj 1. Introduction In a textbook on complex variables we may find real definite integrals of the type I1 = ∞ ∫ 0 dx x2 + 1 or I2 = ∞ ∫ 0 dx x2 (x2 + 1)(x2 + 4) , (1) to be evaluated using the contour integration in the complex plane, making use of the Cauchy’s residue theorem. This is, of course, a simple exercise in complex analysis, and residue theorem is a powerful tool to handle such integrals. However, if we were actually to evaluate them we would do it separately, each integral in its turn, with its own residues of poles summed over to get the final answer. Not so if we use the NDIM technique [1, 2], as we shall shortly see. In NDIM we can integrate both integrals at the same (2767) 2768 A.T. Suzuki time, and more, without having to do it one by one separately; we simply need to evaluate the general integral I(m,n,p) g = ∞ ∫ 0 dx (x2)m (x2 + 1)n(x2 + 4)p (2) and then work out the result for a particular set of numbers m,n, p = 0, 1, 2, 3, . . . , of which I1 and I2 are specific examples. The concept of negative dimensional integration can be best seen in the following D-dimensional Gaussian integration: G = ∫ dDx e−αx2 = ( √ π α )D . (3) On the other hand, expanding in power series the exponential function in the integrand of (3) above, we have G = ∫ dDx e−αx2 = ∫ dDx ∞ ∑ n=0 (−)n αn n! (x2)n = ∞ ∑ n=0 (−1)n αn n! ∫ dDx(x2)n . (4) Now, comparing (3) and (4), we conclude that for the equality to hold we must have I ≡ ∫ dDx(x2)n = (−1)−nn! ( √ π)D δn+ D 2 ,0 (5) and since, by construction n ≥ 0, one is led to assume that D is a negative valued dimension. This non-trivial result for a D-dimensional integral with pure quadratic integrand elevated to a given power was first pointed out by Ricotta and Hal- liday in [3], which differed somewhat from all the previous considerations, starting from dimensional regularization scheme developed by ’t Hooft and Veltman [4], where all such integrals were set to zero straightforwardly. Since Feynman loop diagrams lead to Feynman integrals in D-dimensions in the dimensional regularization scheme, integrals evaluated with the help of (5) in negative dimensions, must be brought back by analytic continuation to the realm of positive dimensions D. Since the concept of negative dimensional Evaluating Residues and Integrals Through . . . 2769 integral arises as a consequence of the positiveness of the polynomial powers in the integrands, the mentioned analytic continuation is achieved by em- ploying a property of the Pocchammer’s symbols bearing these parameters labelling powers of the polynomial expressions in the integrands, namely [5] (a)−k = (−)k (1 − a)k , (6) where (a)k = Γ (a + k) Γ (a) . Of course, here we strict ourselves to one-dimensional integrals such as (2), so that (5) becomes ID=1 ≡ ∫ dx(x2)n = (−1)−nn! √ π δn+ 1 2 ,0 (7) and all the NDIM formulation developed for D-dimensional Feynman integ- rals can be applied to evaluate one-dimensional integrals of the type I1 and I2. 2. Review of Cauchy’s residue theorem application Consider, for practical example, the integral I1 = ∞ ∫ 0 dx x2 + 1 = 1 2 +∞ ∫ −∞ dx x2 + 1 . (8) The second integral on the r.h.s. of the above represents an integration along the real axis of the function f(z) = 1 z2 + 1 = 1 (z + i)(z − i) , (9) which is a function of complex variables with simple poles at z = ±i. Let CR be the semi-circle with radius |z| = R with R > 1 as shown in figure below - Re(x) 6 Im(x) q-i qi −R R - - I CR 2770 A.T. Suzuki By Cauchy’s theorem, we have R ∫ −R dx f(x) + ∫ CR dz f(z) = 2πi κ1 , (10) where κ1 is the residue of f(z) at the (simple) pole z = i, which, of course, can be calculated easily by κ1 = (z − i)f(z)|z=i = 1 z + i ∣ ∣ ∣ ∣ z=i = 1 2i . (11) Therefore, for R > 1, we have from (10) R ∫ −R dx x2 + 1 = π − ∫ CR dz z2 + 1 . (12) Now, |z| = R when z is on the semi-circle CR, so that |z2 + 1| ≥ |z|2 − 1 = R2 − 1 , (13) and therefore ∫ CR dz z2 + 1 ≤ ∫ CR |dz| R2 − 1 = πR R2 − 1 → 0 for R → ∞ . (14) Then, taking the limit lim R→∞ R ∫ −R dx x2 + 1 = +∞ ∫ −∞ dx x2 + 1 = π , (15) or I1 = 1 2 +∞ ∫ −∞ dx x2 + 1 = π 2 . (16) 3. Negative dimension integration method On the other hand, let us take the Gaussian generating funtional of the negative dimensional integration, namely, I = +∞ ∫ −∞ dx e−α(x2+1) , (17) Evaluating Residues and Integrals Through . . . 2771 where we note that the argument in the exponential is the denominator of the integrand in (8) multiplied by a real, positive parameter α which is chosen as a converging factor for the integral. Of course, this is the standard Gaussian integral, whose result is I = e−απ 1 2 α− 1 2 = π 1 2 ∞ ∑ j=0 (−)j (α)j− 1 2 j! . (18) If we project out in power series the integrand of (17) before the inte- gration is done, we have I = ∞ ∑ n=0 (−)n αn n! INDIM(n) , (19) where the negative dimensional integral INDIM(n) is given by INDIM(n) ≡ +∞ ∫ −∞ dx (x2 + 1)n . (20) Note that for n = −1 this is exactly the integral (8) we want to evaluate. Observe, however, that the negative dimensional integral is more general than the one we have in (8) in the sense that the exponent n is not fixed. Note however that (20) is only defined for positive n, so that in order to get the result for (8) we need to make an analytic continuation to negative values of n, a process whereby the integral gets defined into positive dimensionality in (20). Comparing the two series expansion (18) with (19), we observe that in order to both series be equivalent, we must have n = j − 1 2 , so that INDIM(n) ≡ +∞ ∫ −∞ dx (x2 + 1)n = (−)−n n!π 1 2 (−)n+ 1 2 (n + 1 2)! = (−π) 1 2 Γ (n + 1) Γ (n + 1 2 + 1) = (−π) 1 2 (n + 1) 1 2 . (21) Now, analytic continuing (AC) the exponent n into negative values using the property (6), we have I (AC) NDIM(n) = +∞ ∫ −∞ dx (x2 + 1)−n = π 1 2 (−n) − 1 2 = π 1 2 Γ (−n − 1 2) Γ (−n) . (22) 2772 A.T. Suzuki Now, substituting n = −1,−2,−3, . . . , we get I (AC) NDIM(−1) = +∞ ∫ −∞ dx x2 + 1 = π , (23) I (AC) NDIM(−2) = +∞ ∫ −∞ dx (x2 + 1)2 = π 2 , (24) I (AC) NDIM(−3) = +∞ ∫ −∞ dx (x2 + 1)3 = 3π 8 , (25) ... = ... = ... , and so on and so forth. Of course, for the original integrals, which range from [0,∞), the corresponding values are half of the values quoted above. A more interesting case is the evaluation of the other integral, namely, I2 = ∞ ∫ 0 dx x2 (x2 + 1)(x2 + 4) = 1 2 +∞ ∫ −∞ dx x2 (x2 + 1)(x2 + 4) . (26) Evaluation of this integral by the residue technique is similar to the previous one where now we have two poles in the upper hemisphere, z = i and z = 2i, so that the contour CR is such that its radius must be |z|=R, R > 2, before taking the limit R → ∞. The result, after summing the residues of both poles is I2 = 1 2 2πi { − 1 6i + 1 3i } = 1 2 { −π 3 + 2π 3 } = π 6 . (27) We have written down the explicit contributions of each residue, the first one corresponding to the residue at the simple pole z = i and the second one to the residue at z = 2i, to show that these residues correlate with each of the “basis” (linearly independent) solutions (with a word borrowed from the language of basis vectors in a vector space) in NDIM. Evaluating Residues and Integrals Through . . . 2773 The Gaussian generating functional of the negative dimensional integral to this case is J ≡ +∞ ∫ −∞ dx e−αx2 −β(x2+1)−γ(x2+4) = e−β−4γ +∞ ∫ −∞ dx e−(α+β+γ)x2 = e−β−4γ π 1 2 (α + β + γ) 1 2 = π 1 2 (α + β + γ) 1 2 ∞ ∑ r,s=0 (−)r+s βr r! 4s γs s! = π 1 2 Γ ( 1 2 ) a+b+c=− 1 2 ∑ r,s,a,b,c (−)r+s4s αaβb+rγc+s a! b! c! r! s! , (28) where in the last line of the above we have employed the standard multino- mial expansion for (α + β + γ)− 1 2 . On the other hand, direct expansion in power series of the integrand yields J = ∞ ∑ j,l,m=0 (−)j+l+m αjβlγm j! l!m! +∞ ∫ −∞ dx (x2)j(x2 + 1)l(x2 + 4)m = ∞ ∑ j,l,m=0 (−)j+l+m αjβlγm j! l!m! INDIM(j, l,m) . (29) Comparing (28) with (29) we see that we must have j = a , l = b + r , m = c + s , a + b + c = −1 2 . (30) Also, when we compare the two series, the one in (28) has five summation indices with one constraint, whereas the other one (29) has three summa- tion indices. Then our result for INDIM(, j, l,m) will be given in terms of 5−3−1 = 1, that is, a single summation index. However, since we have five indices in one and three in the other, with one constraint equation among them, one can in principle have the following combinatorial possibilities: 2774 A.T. Suzuki 5C4 = 5! 4! 1! = 5 (31) for the remaining series index. Letting r be the summation index in the result, we have then the follow- ing conditions: a = j , b = l − r , c = −j − l − 1 2 + r , s = j + l + m + 1 2 − r , (32) so that the negative dimensional integral will be given by I (r) NDIM(j, l,m) = (−)−j−l−mj! l!m!π 1 2 Γ (1 2) × ∞ ∑ r=0 (−)j+l+m+ 1 2 4 j+l+m+ 1 2 −r j! (l − r)! (−j − l − 1 2 + r)! r! (j + l + m + 1 2 − r)! = (−π) 1 2 4 j+l+m+ 1 2 Γ (1 2)Γ (1 + m) Γ (1 − j − l − 1 2 )Γ (1 + j + l + m + 1 2) × ∞ ∑ r=0 1 (1 + l)−r(1 − j − l − 1 2)r(1 + j + l + m + 1 2)−r4rr! = (−π) 1 2 4j+l+m+ 1 2 (1 2 )−j−l (1 + m)j+l+ 1 2 ∞ ∑ r=0 (−l)r (−j − l − m − 1 2)r (1 − j − l − 1 2)r (1 4 )r r! = (−π) 1 2 4j+l+m+ 1 2 (1 2 )−j−l (1 + m)j+l+ 1 2 2 ×F1 ( −l , −j − l − m − 1 2 1 − j − l − 1 2 ∣ ∣ ∣ ∣ ∣ 1 4 ) . (33) Analogous calculations can be performed for s, b, c indices, whereas for the index a there is no solution except the trivial one (this is the case when the determinant of the system of linear equations vanishes). For the remain- ing non-vanishing determinants, we have then the solutions: For s remaining summation index, we have the conditions Evaluating Residues and Integrals Through . . . 2775 a = j , b = −j − m − 1 2 + s , c = −m − s , r = j + l + m + 1 2 − s , (34) and the solution yields: I (s) NDIM(j, l,m) = (−π) 1 2 (1 2 )−j−m(1 + l)j+m+ 1 2 ∞ ∑ s=0 (−m)s(−j −l −m − 1 2)s (1− j− m− 1 2)r 4s s! = (−π) 1 2 (1 2 )−j−m (1 + l)j+m+ 1 2 2 F1 ( −m ,−j −l −m − 1 2 1 − j − m − 1 2 ∣ ∣ ∣ ∣ ∣ 4 ) . (35) For b summation, we have a = j , r = l − b , s = j + m + 1 2 + b , c = −j − 1 2 − b , (36) and the solution yields: I (b) NDIM(j, l,m) = (−π) 1 2 4j+m+ 1 2 (1 2)−j (1 + m)j+ 1 2 2 F1 ( j + 1 2 , −l 1 + j + m + 1 2 ∣ ∣ ∣ ∣ ∣ 4 ) . (37) Finally, for c summation we have a = j , r = j + l + 1 2 + c , s = m − c , b = −j − 1 2 − c , (38) and the solution yields: I (c) NDIM(j, l,m) = (−π) 1 2 4m (1 2 )−j (1 + l)j+ 1 2 2 F1 ( j + 1 2 , −m 1 + j + l + 1 2 ∣ ∣ ∣ ∣ ∣ 1 4 ) . (39) 2776 A.T. Suzuki Now, observe that we have four non-vanishing solutions arising from the solving of systems of linear equations, which here clearly come in pairs when we look at the argument of the hypergeometric functions in the results. The solution for the integral is then the sum of the pairs with the same argument (linear combination), namely, INDIM(j, l,m) = A 2F1(a, b; c|z−1) + B 2F1(d, e; f |z−1) = A′ 2F1(a ′, b′; c′|z) + B′ 2F1(d ′, e′; f ′|z) , (40) where the coefficients A, B, A′ and B′ are given by A = (−π) 1 2 4j+l+m+ 1 2 (1 2)−j−l (1 + m)j+l+ 1 2 , B = (−π) 1 2 4m (1 2)−j (1 + l)j+ 1 2 , A′ = (−π) 1 2 (1 2)−j−m (1 + l)j+m+ 1 2 , B′ = (−π) 1 2 4j+m+ 1 2 (1 2)−j (1 + m)j+ 1 2 , (41) and the hypergeometric function parameters and variables are: a b c d e f z−1 −l −j − l − m − 1 2 1 − j − l − 1 2 −m j + 1 2 1 + j + l + 1 2 1 4 a′ b ′ c′ d ′ e′ f ′ z −m −j − l − m − 1 2 1 − j − m − 1 2 −l j + 1 2 1 + j + m + 1 2 4 Before proceeding, let us demonstrate that the two sets of solutions, namely the primed and unprimed ones are totally equivalent. To do this, we employ the following analytic continuation property of hypergeometric functions [7] 2F1(a, b; c|z) = z−a (1 − b)a (c)−a 2F1(a, 1 + a − c; 1 + a − b|z−1) + z−b (1 − a)b (c)−b 2F1(b, 1 + b − c; 1 + b − a|z−1) , where | arg(−z)| < π , (42) Evaluating Residues and Integrals Through . . . 2777 to one of the basis solutions, say of the unprimed set 2F1 ( −l, −j − l − m − 1 2 ; 1 − j − l − 1 2 | 1 4 ) to get the primed result. Of course, we could have used the transformation property above to the other basis solution 2F1(−m, j + 1 2 ; 1 + j + l + 1 2 |14), and we would get the same primed result. We need to apply only to one of the basis solutions, since the transformation property above referred to cannot produce neither new nor any more than two linearly independent hypergeometric functions. Now, we need to analytic continue the results to negative values of ex- ponents and positive dimension. Observe that the result for INDIM contains two factors: One is the coefficients, given by ratios of gamma functions and the other is the functional part, given by the hypergeometric functions. For the coefficients, which contain ratios of gamma functions given in terms of Pocchhammers symbols, we employ (6), and for the functional part, just let the exponents go to negative valued parameters. Note that we need to be aware of which exponent should be continued to negative values [6]. Then, our final result for the integral reads: IAC NDIM(j, l,m) = AAC 2F1(a, b; c|z−1) + BAC 2F1(d, e; f |z−1) = A′AC 2F1(a ′, b ′; c′|z) + B′AC 2F1(d ′, e′; f ′|z) , (43) where AAC = π 1 2 4j+l+m+ 1 2 ( 1 2 ) j+l (−m) −j−l− 1 2 , BAC = π 1 2 4m ( 1 2 ) j (−l) −j− 1 2 , A′AC = π 1 2 ( 1 2 ) j+m (−l) −j−m− 1 2 , B′AC = π 1 2 4j+m+ 1 2 ( 1 2 ) j (−m) −j− 1 2 . (44) One interesting thing about the NDIM technology is that it allows us to write the correct answer in as many equivalent ways as it is possible to do. For the case in question, we have two equivalent answers, namely, the unprimed and primed answers. So, for particular values of the exponents, say, j = 1 and l = m = −1, which is the case for the I2 integral mentioned in the introduction, we have: 2778 A.T. Suzuki IAC NDIM(1,−1,−1) = π 1 2 { 4− 1 2 (1) − 1 2 2F1 ( 1, 1 2 ; 1 2 |14 ) +4−1 (1 2 )1(1)− 3 2 2F1 ( 1, 3 2 ; 3 2 |14 ) } (45) = π 1 2 { (1) − 1 2 2F1 ( 1, 1 2 ; 1 2 |4 ) + 4 1 2 (1 2)1 (1) − 3 2 2F1 ( 1, 3 2 ; 3 2 |4 ) } . (46) Finally, using the fact that [7] 2F1(a, b; b|z) = (1 − z)−a (47) the two results (45) and (46) coalesce into one: IAC NDIM(1,−1,−1) = −π 3 + 2π 3 = π 3 , (48) so that each of the basis solution corresponds exactly to the residue of the poles at z = i and z = 2i. Finally, I2 = 1 2 IAC NDIM(1,−1,−1) = π 6 . (49) Other particular cases, such as j = m = 0, l = −1, j = 0, l = m = −1 and j = 0, l = −1, m = −2 can be calculated from the general solution, yielding I1 = 1 2 IAC NDIM(0,−1, 0) = π 2 , I (0,1,1) 2 = 1 2 IAC NDIM(0,−1,−1) = π 12 , I (0,1,2) 2 = 1 2 IAC NDIM(0,−1,−2) = 5π 288 . Note that for all the cases where either l = 0 or m = 0 the general solution is such that one of the terms in the overall result vanishes because we have a term proportional to 1/Γ (0) = 0, and the answer is then given by a single hypergeometric function. 4. Conclusions Using the NDIM technique, we evaluated some sample real definite inte- grals which may be calculated by the Cauchy residue theorem in the complex plane. The alternative methodology here presented gives us the bonus in that all the generic exponents of integrands can be calculated at once, from where Evaluating Residues and Integrals Through . . . 2779 particular solutions can be drawn. There is no difficulty in the performing of the integration since the integration involved is of the Gaussian type and the technique requires only series comparison term by term and the solving of systems of algebraic linear equations resulting from such a comparison. We showed that the results for different variables are obtained simultane- ously and they are equivalent to each other. Moreover, for each set of basis solutions correspond the residue of a given pole. The strength of this new technique also can be envisaged in that simple, double or higher order poles can be evaluated all at once. The author gratefully acknowledges the kind hospitality of the Depart- ment of Physics, North Carolina State University, and financial support from CAPES (Brasília). REFERENCES [1] A.T. Suzuki, E.S. Santos, A.G.M. Schmidt, J. Phys. A36, 11859 (2003). [2] G.V. Dunne, I.G. Halliday, Phys. Lett. B193, 247 (1987). [3] R.M. Ricotta, I.G. Halliday, Phys. Lett. B193, 241 (1987). [4] G. ’t Hooft, M. Veltman, Diagrammar CERN report 73-9 (1973). [5] A.T. Suzuki, A.G.M. Schmidt, J. High Energy Phys. 09, 002 (1977). [6] A.T. Suzuki, A.G.M. Schmidt, Eur. Phys. J. C10, 357 (1999). [7] Ed. A. Erdérlyi, Higher Transcendental Functions, Bateman Manuscript Project, Caltech, vol. I, McGraw-Hill Book Company, Inc., 1953.