# Full text of "Calculation of the phi^4 6-loop non-zeta transcendental"

## See other formats

OS 23 Calculation of the </? 4 6-loop non-zeta transcendental Oliver Schnetz December 16, 1999 Abstract ^ . We present an analytic calculation of the first transcendental in y> 4 -Theory that is not of the form £(2n+l). ft is encountered at 6 loops and known to be a weight 8 double sum. Here it is obtained by reducing multiple zeta values of depth < 4. We give a closed expression in ^sO ' terms of a zeta-related sum for a family of diagrams that entails a class of physical graphs. We confirm that this class produces multiple zeta values of weights equal to the crossing numbers of the related knots. ; Contents H Introduction Notation and definitions ^ [3| Calculation of the n-banana diagram 1 Introduction D . 4 Conclusions > X The miraculous connection between Feynman graphs, knots, and numbers discovered by D. Kreimer and D.J. Broadhurst 0-||] reveals a structure underlying generic renormalizable Quan- tum Field Theories. It is conjectured that via Knot Theory the topology of an amplitude pro- vides its transcendental number contents which for primitively divergent diagrams, is given by the renormalization scheme independent numerical factor in front of the logarithmic singularity. The calculation of these numbers was acchieved in ^ for 56 of the 59 primitively divergent graphs up to 7 loops in massless </? 4 -Theory. Diagrams that could not be calculated analytically were evaluated numerically to high precision. A systematic search for the rational coefficients of a conjectured basis of transcendentals led in all but three 7-loop diagrams (related to the hyperbolic knots IO139 an d IO152) to (with very high probability) exact results. The first graphs where analytic calculations were not available showed up at 6 loops. These diagrams entailed the 4,3-torus knot (as opposed to the (2n+l),2-torus knots leading to £(2n+l)) which was determined to be related to the transcendental ^(— £(5, 3) + 29£(8)), where £(n, m) = *Institut fur theoretische Physik III, Staudtstrafie 7, 91058 Erlangen, Germany e-mail: schnetz@theorie3.physik.uni-erlangen.de 1 J2k>e>i k' n £- m . In the terminology of f| these graphs are the M(2, 1, 1, 0) and the M(l, 1, 1, 1) '4-banana' diagrams. In this paper we focus on the more symmetric M(l, 1, 1, 1). We keep equations general as far as this does not complicate calculations. In Theorem 10 a closed result is given that easily generalizes to encorporate all n-banana diagrams (defined below). The derivation is organized as a collection of propositions that are all proved in full detail although most of them are elementary. 2 Notation and definitions We normalize 4-dimensional integrals with -X^ and 2-dimensional integrals with ^ (which is symmetric under Fourier transformation), dx = [ d A x , [ dx = — / d 2 x . (1) In order to handle the general case most efficiently, it is convenient to use a <5-symbol (also 5, 5i, etc.) with the meaning that f(5) is the constant term in the Laurant expansion of / at 5 = 0, /W = /o = ivT dx* xNf{x) ' if/(x)= E fnXn - (2) x=0 n=~N Expample: -ri* = — ln fc (x) , if > (and zero otherwise). Definition 1. For x,y G M let fk(x,y) be recursively defined by fo(x,y) = - — —2- (3) (x-y) 1 1 (x-z) 2 ~ 2 fk+i(x,y) = I dz- - T —f k {z,y) (4) Remark 2. The funcion fk(x,y) depends only on \x\, \y\, and the angle between x and y. We may treat it as a function of two complex variables xc and yc that coincide with x and y if the complex plane is considered as the subspace of M 4 spanned by x and y. Instead of fk(xc, lc) we a ^ so write fk(x,x). We have fk{x,y) = fk{y,x) , f k (x c ,yc) = r^Tof {xc/ycAc) = r^-Tof (yc/xcAc) ■ (5) |yc| pel Definition 3 (n-banana). Let k = (k\, . . ., k n ) be a vector of non-negative integers with not more than one of the fcj = 0. Let —2 < a < n — 2 and e\ be a 4-dimensional unit-vector. Then /n dx(x 2 Yl[f ki (x,ei) . (6) i=\ Note, that the integral is independent of e± and converges under the above conditions. In the notation of (l| we have G (ki, k 2 , k 3 ) = B {h,k 2 , fc 3 ) = B -i (h, kz, h) and M (k x , k 2 , fa, k 4 ) = B (k x , fa, fa, fe 4 ) . 2 3 Calculation of the n-banana diagram We restrict ourselves to— 1 < a < n — 3 and ki > 1 Vi which includes Bq(1, 1, 1, 1). Proposition 4. Let / : R 4 — > R, x i— > /(x) be a function of |x| and x • only, then dxf (x) = -- / dx(x - x) /c (x, x) , 4 Jc where fc : C — > R, x i— > /c(x, x) equals / on the complex plane spanned by x and ei = lc- Proof. Introducing polar coordinates yields (7) dxf (x) 1 47T 2 Jo dxx 47T dtfsin 2 tf/cfse** a?e~ W ) 1 2^ dxx / dd 7T (xe" 9 — xe (xe^xe^) dx (x - x) fc (x, x) . □ Hence, If B a (k) = -- I dx (xx) a (x - x) 2 TT /fc. (x, x) 4 7c ^ With Eq. (|5|) we immediately obtain B a (k) = B n „ 4 -a (k) . Moreover, with x = |x|x, x = |x|/x we get B a (k) dxx 2a+l dx dB 2nix xx — x /xfnfh (xx,x/x) (8) (9) i=l where we integrate x along the unit circle dB in C. Proposition 5. fk (x, x) <5 + £ dtt d — 1 1 xx/ (£ — x) (t — x) (10) With (||) and our definition of the 5-symbol we get the following 4-dimensional statement. Corollary 6. fk (x, y) 4 k kl (k I^/>>- *(£)<^ xY (ii) Proof of Prop. 5. Expansion of the propagator into a sum of Gegenbauer polynomials gives (with cos(& xy ) = x ■ y/\xy\ and x< = x if x < 1, x< = 1/x if x > 1) |xy| (x - y) z x n f dP 22 C n _l (COS (& x y)) - = I 22 nC n ~l (COS (tixy)) n=i y < ra= i 1 (n 2 + P 2 ) (12) Orthogonality of the Gegenbauer Polynomials yields /dP — V nC n -i (cos (i? xei )) n=l |iP-l (n 2 + P 2 ) fe+1 (13) 3 from which we derive the recursion relation d , , , , f dP ^ „ , , „ N x , ,jp_i —id 1 f dP \x\fk (X, ei) = / — nC n-l (COS (fi xei )) \x k d\ X y — ' - J vr ^ v 1 2kdP( n 2 + P 2 ) = — T / / TlC n -l (COS (t^e, )) r — — = : ffc_l(X,ei). 2fc i 7T ^ V V XBlJJ ( n 2 + p2)fe «9P 2fc V ^ If we integrate this relation we encounter an ambiguity of the form c(x)/|x| which is removed by the condition lim^i^,^ |x|/(x,ei) = 0. Equation ( |i~0| ) meets this condition. Moreover with g k (x,x) = J^dt t 5 (t/(xx)) S (l/(t - x) - l/(t - x)) we get 2k d . . 6 + 6 1 , \x\- — — r - g k (x,x) In (\x\) d\x\ (466) x — x 2k 6 + 6 1 ( 25 , s f 1 , x x" 1 ln(|x|)(4M) fe V |x| 5fc(X ' S) + i d " 5 (xx; V(*"^) 2 (t-S)V>/ h ' ^-2(5g fc (g, a) + (- 1 ^ - + Y"T^) ( x ^)~ g + (£ + £) gfc (x, x)\ hi (M) (4(5(5) x ~ x 2k f 1 ( ln k (t/xx) ln k - 2 (t) ln fc - 2 (t/xx) ln k [t) 4 fc ln(|x|)(x-x) Jo \ k\ (fc-2)! (Jfe-2)! fc! J \t - x t-x if jfc > 2 and — — if jfe = 1. 1 — x 1 — x According to Cor. 6 this equals A_i(x,x) which completes the proof. □ We find the following behaviour of fk(xx,x/x) as a function of a complex variable x. Proposition 7. The function fk(xx,x/x) is analytic in x save for two cuts on the real axis ranging from to x< and from l/x< to oo. Moreover, if x = |x|e* £ we have in the limit e — > A (xx, x/x) - / fc (xl, x/1) = 27ft /tA (x| ^!, (x/l fP, ' (e (X< - |X|) - e (|x| - 1/X<)) . v ' (4(5(5) X /M ~~ x Fl (14) Proof. /* (xx,x/x) - f k (xx,x/x S + S f\. s ( t \ S ( f 1 I \ ! , 1 dtt° ^ - — - - - — = - — - + (A66) k Jo \x 2 J y\t - xx t — xx/t — x/x t-xx\t-x/x t — x/x J J (In the limit e — > the poles in the differences move to the points t = x|x| and t = x/\x\ on the real axis. In the case that one of them falls into the interval [0,1] we deform the t-integrals to closed curves C x \$\ and C^/ui encircling the singularities with positive or negative orientation.) ' + 6 T I e (l - x\x\) J c dt + e (i - x/|x|) J c i dt\ t s (JL (4(5(5) \ Jc xm Jc x }\z t — x x t — x/ x 2vri 6 + 6 1 (A66) k x/\x\ — x\x\ ^0 (1 — x|x|) (xlxl)* 5 (x/|x|) 6 + (1 — x/|x|) (x[x|) S (x/\x\) S ^j Changing 6 to — 6 and <5 to — <5 in the second term leads to the claimed result. □ 4 Since, for n > 3 the integrand in Eq. (||) has no pole at x = we may deform the integration contour to range from to e~ ie and back from e ie to (in the limit e — > 0). UX 2TTf - /-\ \ I J* i J- \ \ XX ~ X I X ) r ,L I re re \ XIX) J dB 2^x {X£ ~ x/£)2 g A, = ^ " I dxj — g /*. */ / 2^ " x/5)2 ( II (^ e_fe > ^ e ") - II (**e fe , (x/x) e— ) J - T -^r - ^) 2 E I] /* (^ e ~"> (^/^) e ") • (xxe ie , (x/x) e" ie ) - /*, (xxe" i£ , (x/x) e i£ )) JJ /fcj ( xx ^ , (x/x) e~ ie ) - r ? e n /* e_i£ > eie ) 7^t57 ~ x I*D w*!)* 4 (^/i^i)^ ( 15 ) ■ \{f ki {xxe ie ,(x/x)e~ where n»<i = Ili>n = 1 - Proposition 8. Proof. roo rx< fl x I rl roc / dx x —F(xx,x/x) = - dx dxF(x,x) . (16) Jo Jo x 2 Jo Ji i r% dx f°° f x l x dx dx x — F(xx,x/x) + / dx x — F(xx,x/x) o Jo x Ji Jo x rl rl dx ( \ r°° dx / J dx x J —F\^x 2 x,l/x)+ J dx x J —F[x,x 2 /x rl dx . dx / —F(x,x/x) 2 Jo Ji x 2 J\ Jo x 1 I roo /• rl roo \ - I / dx / dx + dx dx ] F (x, x) 2 y7i Jo Jo Ji/x / - / dx / dx (0 (1/x — x) + (x — 1/x)) F (x,x) = - dx I dx F (x, x) 2 Jo Ji 2 Jo 1 Z" 1 Z" 00 dx 1 f c - dx —F(x/x,x) + - 2 Jo Ji x 2 Ji With the shorthand (4<W) Jo \xxj \t-x t-x we get from Eqs. ^j, ([15]) (for — 1 < a < n — 3 all integrals converge) 1 - (5^ + 8e f 1 f°° x a+5 e ^a-S e b « w = o e 7^7717 L dx L dx j—-^2 n 91 fa x ) n »j ^) • □ 5 Proposition 9. For x < 1 and x > 1 (4M) fc \hi n + 5 + 8 A _l_ A / 00 (5 I ™n+<5„<5 _ \ & *) = TT^i £ + ™~ V (" cot (* (* + «)) ± <) • (19) Proof. (i) (ii) ;±»e /j.\<5i oo „i / + \ <5 + n-l 00 _ 1 dt t s -1 - = -V dtt* - ^ = _y x -^— * xx J t — x — t Jo \xx J x n ' n + 5 + 5 n=l v ' n=l , + \ 5 i / /-ooe ±le i-oo \ / + \ 5 i df t° — = / dt- dt\t x 5 x- 5 xx J t — x Wo Jl j \xx J t — x . /•ooe J -' t _ 1 J^> poo / f \ o ~.n i _v / dtt'f — ) ^j,. io t-l \XX) Now, (C" 1 is a negatively orientated circle around x) / dt- dt) t S+6 — = / dt t 5+S — = -2m , Jo Jo J t-l Jc- 1 t-l and assuming (without restriction) 8 + 8 < fy dt + y cftl^-i- = 2Rel-y dtt s+s J2t n ~ 1 + J dtt 5+5 £ t™" 1 n=l 00 1 = -2 V = = -2vrcot(vr(5 + 5)) . n=— 00 Hence, re ±t£ / + \ 5 1 00 „,n— <5™-5 /„ d < (i (5 — " " £ " ^ M (" + ± ■ ' ra=0 The n = 0-term drops for > by definition of the 5-symbol. Changing 8 to —5 and 8 to —5 in (ii) gives the result. □ With (x-x)~ n+2 = J2n e =i ( n ^™ 3 " 4 )^" 1 ^" n< " n+3 the x-, x-integrals in Eq. (||) are trivially evaluated and we obtain the following result. Theorem 10. Let —1 < a < n — 3 and ki > 1 VI With a = n — 4 — a and OfF{P j J j )= f] ^ ( ~* J ' WJ ± ^ ^ ^ + ^' =^ +vrF (-^, - J 3 -) (-oot^^ + ^ii) (20) we obtain (cf. Def. 3) «. « = £ n ^ £ £ (' ! ';: 3 " 4 ) n ^ n ( A jj) , <«) with F ^) = _i_ r ^— zx— ^ - • ( 22 ) □ 6 Cj = - ; - (23) In the case a = a there exists a symmetry interchanging bared with unbared variables. This allows us to reduce the case n = 4, ki = k Vi to 6 sums. With the definitions Aj = "v^ i v^T ^T~i — ' Aj = ^ , ^/ , for J C {1, ...,»}, E n j + E Oj - Edj + a E n i + E °j - E Oj + a jeJ jeJ j^j jeJ jeJ j^j 1 rij + Sj + S we find n r , ? oo oo B (k,k,k,k) = II - * ( £ SAG^l^l + 3 £ C 2 C 3 C 4 A 123 A U i=l (4<5i<5j) 711,712,713,714=1 77,1,712, 713,714=1 (00 00 £ C2C 3 A 123 ii+ £ C2C 3 yli 2 li3 711,712,713 = 1 721,712,713 = 1 OO + vr 2 (3 cot (vr (S 3 + 5 3 )) cot (vr (<5 4 + 5 4 )) ~ £ C 2 Ai 2 Ai (24) 711,712 = 1 F ■ cot (vr (<5 2 + <5 2 )) (cot (vr (<5 3 + 5 3 )) cot (vr (5 A + l 4 )) — 1) ■ ~ 711=1 Each of the sums is convergent. The last sum is easily evaluated to be a multiple of vr 8 . To convert sums 1, 3, 5 into multiple zeta values it is sufficient to make repeated use of the identity 11 1 / 1 1 \ + — ( 25 ) 7T Hi n 2 m + n 2 \ni n 2/ either before of after the expansion of the 5's. Sums 2, 4 are first to be converted into partial fractions with respect to fc 4 or k 3 , respectively. Then one uses the identity (i = 4, 3) oo oo oo oo £ F(k i + k 1 ,k 1 )= £ F(h,h)- £ F(fci,fca + fci)- £ (26) fci,fei=l fei,fci=l fei,fej=l fci=l and (|2^) to produce multiple zeta values. Finally we use tables (due to D.J. Broadhurst) to convert the multiple zeta values of weight 8 to a minimum basis consisting of £(5,3), C(8), £(5)£(3), and C(3) 2 £(2). We obtain * (1,1,1,1) - -^(^) + ^fc( 8 )-^C(5,C( 3) + |c( 3 ) 2 C( 2 ) Bl( (5, 3) - i^C W + (5) < (3) - °° C (3) 2 C (2) 320 v y 3840 w 64 s w s w 64 s v ; s v ; 135 , . 14555 , , 180 w N w x 18 , s2 . x 32o c(5 - 3 ) + i84(r c(8 >-isr <;(5K(3) -64 c(3) c(2) 640 ; c( 8 ) 3840 7 27 2fi1 81 B (1, 1, 1, 1) = --C (5, 3) + — C (8) - -C (5) C (3) (27) which reproduces the result in Q up to a factor of 4 5 which is due to a different normalization of the integrals. 4 Conclusions We presented a closed expression for n-banana diagrams in terms of sums that easily convert to multiple zeta values. The derivation is valid for — 1 < a < n— 3 whereas the n-banana is defined for —2 < a < n—2. If we approach the limit a = -1 or a = n - 3 in Eq. (|2l] ) we encounter terms of the form l/0 k for k > 0. The limit, however, is finite. We conclude that the coefficients in front of the singular terms add up to zero. We thus expect Eq. fl2l|) to be valid in the region — 2 < a < n — 2 if one nullifies singular terms. The inclusion of the case ki = for one i G {0, . . .,n} is straight forward and amouts to basically replacing the numerators <5j + <5j by linear combinations of the n^-'s. It is hence possible to treat all n-banana diagrams with our method. Analogous results may be derived in any even dimensions. Equation ( ^l"l) can be used to study the cases where level-mixing does not occur. If n > 5 we expect in general level-mixing due to the inhomogeneous factor of ( ne ^^ 4 ) ■ Prom Eq. (||) we read off that for ki > 1 the 4-banana in homogeneous of weight 2(k± + k2 + &3 + k^). We expect the same behaviour if one of the k^s is zero. This confirms the conjecture in Q that the class Bo(k, 1, 1, 1) of graphs related to renormalizable Quantum Field Theories reduces to multiple zeta values of weight equal to the crossing number of the related knots. The cases where a = —1, a = n — 2, or n = 3 are more subtle because of the occurence of the singular terms. These cases include a second physical series Bo(m + n + 2, k, 0) which is conjectured to lead to weight 2(n + m + k) + 5 multiple zeta transcendentals. These together with two more classes of non-banana diagrams are expected to exhaust all independent multiple zeta values up to weight 16 (corresponding to 9 loops). We would not be surprised if future work reveals a simpler derivation of f?o(l, 1, 1, 1) since jBo(1, 1,1,1) seems genuinly to be a two-fold sum: The coefficients in front of the £(3) 2 £(2)- transcendental add up to zero. Since the reduction of weight 8 depth 2 multiple zeta values (in contrast to depth > 2) never produces £(3) 2 C(2)-terms the miracle of the vanishing coefficient could be naturally avoided if there was a calculation of Bq(1, 1, 1, 1) that does not make use of depth > 2. Moreover our calculation does not explain the connection to the (4,3)-torus-knot (and the Ts^Ta^-factor knot). We close the paper with the conjecture that any amplitude reduces to multiple zeta values if it has an angular graph that reduces to a single line (• — •) after repeated substitutions of double lines (•==•) and iterated lines (• — • — •) by single lines. Those amplitudes are free of 6- or higher-j-symbols. They do not entail the full structure of 4 dimensions and are therefore accessible to reducing 4-dimensional integrals to integrals over the complex plane which lies in the heart of our calculations. References 8 [1] D.J. Broadhurst, D. Kreimer, Knots and Numbers in cf) 4 Theory to 7 Loops and Beyond, [hep-ph/9504352| . [2] D.J. Broadhurst, D. Kreimer, Association of multiple zeta values with positive knots via Feynman diagrams up to 9 loops, hep-th/9609128 . [3] D. Kreimer, Renormalization and Knot Theory, q-alg/9607022| . [4] D.J. Broadhurst, J. A. Gracey, D. Kreimer Beyond the triangle and uniqueness relations: non-zeta counterterms at large N from positive knots, hep-th/9607174 . [5] D.J. Broadhurst, Solving differential equations for 3-loop diagrams: relation to hyperbolic geometry and knot theory, hep-th/9806174 . 9