Barnes G-function

Plot of the Barnes G function G(z) in the complex plane from -2-2i to 2+2i with colors created with Mathematica 13.1 function ComplexPlot3D
Plot of the Barnes G aka double gamma function G(z) in the complex plane from -2-2i to 2+2i with colors created with Mathematica 13.1 function ComplexPlot3D
The Barnes G function along part of the real axis

In mathematics, the Barnes G-function G(z) is a function that is an extension of superfactorials to the complex numbers. It is related to the gamma function, the K-function and the Glaisher–Kinkelin constant, and was named after mathematician Ernest William Barnes.[1] It can be written in terms of the double gamma function.

Formally, the Barnes G-function is defined in the following Weierstrass product form:

G ( 1 + z ) = ( 2 π ) z / 2 exp ( z + z 2 ( 1 + γ ) 2 ) k = 1 { ( 1 + z k ) k exp ( z 2 2 k z ) } {\displaystyle G(1+z)=(2\pi )^{z/2}\exp \left(-{\frac {z+z^{2}(1+\gamma )}{2}}\right)\,\prod _{k=1}^{\infty }\left\{\left(1+{\frac {z}{k}}\right)^{k}\exp \left({\frac {z^{2}}{2k}}-z\right)\right\}}

where γ {\displaystyle \,\gamma } is the Euler–Mascheroni constant, exp(x) = ex is the exponential function, and Π denotes multiplication (capital pi notation).

The integral representation, which may be deduced from the relation to the double gamma function, is

log G ( 1 + z ) = z 2 log ( 2 π ) + 0 d t t [ 1 e z t 4 sinh 2 t 2 + z 2 2 e t z t ] {\displaystyle \log G(1+z)={\frac {z}{2}}\log(2\pi )+\int _{0}^{\infty }{\frac {dt}{t}}\left[{\frac {1-e^{-zt}}{4\sinh ^{2}{\frac {t}{2}}}}+{\frac {z^{2}}{2}}e^{-t}-{\frac {z}{t}}\right]}

As an entire function, G is of order two, and of infinite type. This can be deduced from the asymptotic expansion given below.

Functional equation and integer arguments

The Barnes G-function satisfies the functional equation

G ( z + 1 ) = Γ ( z ) G ( z ) {\displaystyle G(z+1)=\Gamma (z)\,G(z)}

with normalisation G(1) = 1. Note the similarity between the functional equation of the Barnes G-function and that of the Euler gamma function:

Γ ( z + 1 ) = z Γ ( z ) . {\displaystyle \Gamma (z+1)=z\,\Gamma (z).}

The functional equation implies that G takes the following values at integer arguments:

G ( n ) = { 0 if  n = 0 , 1 , 2 , i = 0 n 2 i ! if  n = 1 , 2 , {\displaystyle G(n)={\begin{cases}0&{\text{if }}n=0,-1,-2,\dots \\\prod _{i=0}^{n-2}i!&{\text{if }}n=1,2,\dots \end{cases}}}

(in particular, G ( 0 ) = 0 , G ( 1 ) = 1 {\displaystyle \,G(0)=0,G(1)=1} ) and thus

G ( n ) = ( Γ ( n ) ) n 1 K ( n ) {\displaystyle G(n)={\frac {(\Gamma (n))^{n-1}}{K(n)}}}

where Γ ( x ) {\displaystyle \,\Gamma (x)} denotes the gamma function and K denotes the K-function. The functional equation uniquely defines the Barnes G-function if the convexity condition,

( x 1 ) d 3 d x 3 log ( G ( x ) ) 0 {\displaystyle (\forall x\geq 1)\,{\frac {\mathrm {d} ^{3}}{\mathrm {d} x^{3}}}\log(G(x))\geq 0}

is added.[2] Additionally, the Barnes G-function satisfies the duplication formula,[3]

G ( x ) G ( x + 1 2 ) 2 G ( x + 1 ) = e 1 4 A 3 2 2 x 2 + 3 x 11 12 π x 1 2 G ( 2 x ) {\displaystyle G(x)G\left(x+{\frac {1}{2}}\right)^{2}G(x+1)=e^{\frac {1}{4}}A^{-3}2^{-2x^{2}+3x-{\frac {11}{12}}}\pi ^{x-{\frac {1}{2}}}G\left(2x\right)} ,

where A {\displaystyle A} is the Glaisher–Kinkelin constant.

Characterisation

Similar to the Bohr–Mollerup theorem for the gamma function, for a constant c > 0 {\displaystyle c>0} , we have for f ( x ) = c G ( x ) {\displaystyle f(x)=cG(x)} [4]

f ( x + 1 ) = Γ ( x ) f ( x ) {\displaystyle f(x+1)=\Gamma (x)f(x)}

and for x > 0 {\displaystyle x>0}

f ( x + n ) Γ ( x ) n n ( x 2 ) f ( n ) {\displaystyle f(x+n)\sim \Gamma (x)^{n}n^{x \choose 2}f(n)}

as n {\displaystyle n\to \infty } .

Reflection formula

The difference equation for the G-function, in conjunction with the functional equation for the gamma function, can be used to obtain the following reflection formula for the Barnes G-function (originally proved by Hermann Kinkelin):

log G ( 1 z ) = log G ( 1 + z ) z log 2 π + 0 z π x cot π x d x . {\displaystyle \log G(1-z)=\log G(1+z)-z\log 2\pi +\int _{0}^{z}\pi x\cot \pi x\,dx.}

The log-tangent integral on the right-hand side can be evaluated in terms of the Clausen function (of order 2), as is shown below:

2 π log ( G ( 1 z ) G ( 1 + z ) ) = 2 π z log ( sin π z π ) + Cl 2 ( 2 π z ) {\displaystyle 2\pi \log \left({\frac {G(1-z)}{G(1+z)}}\right)=2\pi z\log \left({\frac {\sin \pi z}{\pi }}\right)+\operatorname {Cl} _{2}(2\pi z)}

The proof of this result hinges on the following evaluation of the cotangent integral: introducing the notation Lc ( z ) {\displaystyle \operatorname {Lc} (z)} for the log-cotangent integral, and using the fact that ( d / d x ) log ( sin π x ) = π cot π x {\displaystyle \,(d/dx)\log(\sin \pi x)=\pi \cot \pi x} , an integration by parts gives

Lc ( z ) = 0 z π x cot π x d x = z log ( sin π z ) 0 z log ( sin π x ) d x = z log ( sin π z ) 0 z [ log ( 2 sin π x ) log 2 ] d x = z log ( 2 sin π z ) 0 z log ( 2 sin π x ) d x . {\displaystyle {\begin{aligned}\operatorname {Lc} (z)&=\int _{0}^{z}\pi x\cot \pi x\,dx\\&=z\log(\sin \pi z)-\int _{0}^{z}\log(\sin \pi x)\,dx\\&=z\log(\sin \pi z)-\int _{0}^{z}{\Bigg [}\log(2\sin \pi x)-\log 2{\Bigg ]}\,dx\\&=z\log(2\sin \pi z)-\int _{0}^{z}\log(2\sin \pi x)\,dx.\end{aligned}}}

Performing the integral substitution y = 2 π x d x = d y / ( 2 π ) {\displaystyle \,y=2\pi x\Rightarrow dx=dy/(2\pi )} gives

z log ( 2 sin π z ) 1 2 π 0 2 π z log ( 2 sin y 2 ) d y . {\displaystyle z\log(2\sin \pi z)-{\frac {1}{2\pi }}\int _{0}^{2\pi z}\log \left(2\sin {\frac {y}{2}}\right)\,dy.}

The Clausen function – of second order – has the integral representation

Cl 2 ( θ ) = 0 θ log | 2 sin x 2 | d x . {\displaystyle \operatorname {Cl} _{2}(\theta )=-\int _{0}^{\theta }\log {\Bigg |}2\sin {\frac {x}{2}}{\Bigg |}\,dx.}

However, within the interval 0 < θ < 2 π {\displaystyle \,0<\theta <2\pi } , the absolute value sign within the integrand can be omitted, since within the range the 'half-sine' function in the integral is strictly positive, and strictly non-zero. Comparing this definition with the result above for the logtangent integral, the following relation clearly holds:

Lc ( z ) = z log ( 2 sin π z ) + 1 2 π Cl 2 ( 2 π z ) . {\displaystyle \operatorname {Lc} (z)=z\log(2\sin \pi z)+{\frac {1}{2\pi }}\operatorname {Cl} _{2}(2\pi z).}

Thus, after a slight rearrangement of terms, the proof is complete:

2 π log ( G ( 1 z ) G ( 1 + z ) ) = 2 π z log ( sin π z π ) + Cl 2 ( 2 π z ) . {\displaystyle 2\pi \log \left({\frac {G(1-z)}{G(1+z)}}\right)=2\pi z\log \left({\frac {\sin \pi z}{\pi }}\right)+\operatorname {Cl} _{2}(2\pi z)\,.\,\Box }

Using the relation G ( 1 + z ) = Γ ( z ) G ( z ) {\displaystyle \,G(1+z)=\Gamma (z)\,G(z)} and dividing the reflection formula by a factor of 2 π {\displaystyle \,2\pi } gives the equivalent form:

log ( G ( 1 z ) G ( z ) ) = z log ( sin π z π ) + log Γ ( z ) + 1 2 π Cl 2 ( 2 π z ) {\displaystyle \log \left({\frac {G(1-z)}{G(z)}}\right)=z\log \left({\frac {\sin \pi z}{\pi }}\right)+\log \Gamma (z)+{\frac {1}{2\pi }}\operatorname {Cl} _{2}(2\pi z)}

Adamchik (2003) has given an equivalent form of the reflection formula, but with a different proof.[5]

Replacing z with 1/2 − z in the previous reflection formula gives, after some simplification, the equivalent formula shown below (involving Bernoulli polynomials):

log ( G ( 1 2 + z ) G ( 1 2 z ) ) = log Γ ( 1 2 z ) + B 1 ( z ) log 2 π + 1 2 log 2 + π 0 z B 1 ( x ) tan π x d x {\displaystyle \log \left({\frac {G\left({\frac {1}{2}}+z\right)}{G\left({\frac {1}{2}}-z\right)}}\right)=\log \Gamma \left({\frac {1}{2}}-z\right)+B_{1}(z)\log 2\pi +{\frac {1}{2}}\log 2+\pi \int _{0}^{z}B_{1}(x)\tan \pi x\,dx}

Taylor series expansion

By Taylor's theorem, and considering the logarithmic derivatives of the Barnes function, the following series expansion can be obtained:

log G ( 1 + z ) = z 2 log 2 π ( z + ( 1 + γ ) z 2 2 ) + k = 2 ( 1 ) k ζ ( k ) k + 1 z k + 1 . {\displaystyle \log G(1+z)={\frac {z}{2}}\log 2\pi -\left({\frac {z+(1+\gamma )z^{2}}{2}}\right)+\sum _{k=2}^{\infty }(-1)^{k}{\frac {\zeta (k)}{k+1}}z^{k+1}.}

It is valid for 0 < z < 1 {\displaystyle \,0<z<1} . Here, ζ ( x ) {\displaystyle \,\zeta (x)} is the Riemann zeta function:

ζ ( s ) = n = 1 1 n s . {\displaystyle \zeta (s)=\sum _{n=1}^{\infty }{\frac {1}{n^{s}}}.}

Exponentiating both sides of the Taylor expansion gives:

G ( 1 + z ) = exp [ z 2 log 2 π ( z + ( 1 + γ ) z 2 2 ) + k = 2 ( 1 ) k ζ ( k ) k + 1 z k + 1 ] = ( 2 π ) z / 2 exp [ z + ( 1 + γ ) z 2 2 ] exp [ k = 2 ( 1 ) k ζ ( k ) k + 1 z k + 1 ] . {\displaystyle {\begin{aligned}G(1+z)&=\exp \left[{\frac {z}{2}}\log 2\pi -\left({\frac {z+(1+\gamma )z^{2}}{2}}\right)+\sum _{k=2}^{\infty }(-1)^{k}{\frac {\zeta (k)}{k+1}}z^{k+1}\right]\\&=(2\pi )^{z/2}\exp \left[-{\frac {z+(1+\gamma )z^{2}}{2}}\right]\exp \left[\sum _{k=2}^{\infty }(-1)^{k}{\frac {\zeta (k)}{k+1}}z^{k+1}\right].\end{aligned}}}

Comparing this with the Weierstrass product form of the Barnes function gives the following relation:

exp [ k = 2 ( 1 ) k ζ ( k ) k + 1 z k + 1 ] = k = 1 { ( 1 + z k ) k exp ( z 2 2 k z ) } {\displaystyle \exp \left[\sum _{k=2}^{\infty }(-1)^{k}{\frac {\zeta (k)}{k+1}}z^{k+1}\right]=\prod _{k=1}^{\infty }\left\{\left(1+{\frac {z}{k}}\right)^{k}\exp \left({\frac {z^{2}}{2k}}-z\right)\right\}}

Multiplication formula

Like the gamma function, the G-function also has a multiplication formula:[6]

G ( n z ) = K ( n ) n n 2 z 2 / 2 n z ( 2 π ) n 2 n 2 z i = 0 n 1 j = 0 n 1 G ( z + i + j n ) {\displaystyle G(nz)=K(n)n^{n^{2}z^{2}/2-nz}(2\pi )^{-{\frac {n^{2}-n}{2}}z}\prod _{i=0}^{n-1}\prod _{j=0}^{n-1}G\left(z+{\frac {i+j}{n}}\right)}

where K ( n ) {\displaystyle K(n)} is a constant given by:

K ( n ) = e ( n 2 1 ) ζ ( 1 ) n 5 12 ( 2 π ) ( n 1 ) / 2 = ( A e 1 12 ) n 2 1 n 5 12 ( 2 π ) ( n 1 ) / 2 . {\displaystyle K(n)=e^{-(n^{2}-1)\zeta ^{\prime }(-1)}\cdot n^{\frac {5}{12}}\cdot (2\pi )^{(n-1)/2}\,=\,(Ae^{-{\frac {1}{12}}})^{n^{2}-1}\cdot n^{\frac {5}{12}}\cdot (2\pi )^{(n-1)/2}.}

Here ζ {\displaystyle \zeta ^{\prime }} is the derivative of the Riemann zeta function and A {\displaystyle A} is the Glaisher–Kinkelin constant.

Absolute value

It holds true that G ( z ¯ ) = G ( z ) ¯ {\displaystyle G({\overline {z}})={\overline {G(z)}}} , thus | G ( z ) | 2 = G ( z ) G ( z ¯ ) {\displaystyle |G(z)|^{2}=G(z)G({\overline {z}})} . From this relation and by the above presented Weierstrass product form one can show that

| G ( x + i y ) | = | G ( x ) | exp ( y 2 1 + γ 2 ) 1 + y 2 x 2 k = 1 ( 1 + y 2 ( x + k ) 2 ) k + 1 exp ( y 2 k ) . {\displaystyle |G(x+iy)|=|G(x)|\exp \left(y^{2}{\frac {1+\gamma }{2}}\right){\sqrt {1+{\frac {y^{2}}{x^{2}}}}}{\sqrt {\prod _{k=1}^{\infty }\left(1+{\frac {y^{2}}{(x+k)^{2}}}\right)^{k+1}\exp \left(-{\frac {y^{2}}{k}}\right)}}.}

This relation is valid for arbitrary x R { 0 , 1 , 2 , } {\displaystyle x\in \mathbb {R} \setminus \{0,-1,-2,\dots \}} , and y R {\displaystyle y\in \mathbb {R} } . If x = 0 {\displaystyle x=0} , then the below formula is valid instead:

| G ( i y ) | = y exp ( y 2 1 + γ 2 ) k = 1 ( 1 + y 2 k 2 ) k + 1 exp ( y 2 k ) {\displaystyle |G(iy)|=y\exp \left(y^{2}{\frac {1+\gamma }{2}}\right){\sqrt {\prod _{k=1}^{\infty }\left(1+{\frac {y^{2}}{k^{2}}}\right)^{k+1}\exp \left(-{\frac {y^{2}}{k}}\right)}}}

for arbitrary real y.

Asymptotic expansion

The logarithm of G(z + 1) has the following asymptotic expansion, as established by Barnes:

log G ( z + 1 ) = z 2 2 log z 3 z 2 4 + z 2 log 2 π 1 12 log z + ( 1 12 log A ) + k = 1 N B 2 k + 2 4 k ( k + 1 ) z 2 k   +   O ( 1 z 2 N + 2 ) . {\displaystyle {\begin{aligned}\log G(z+1)={}&{\frac {z^{2}}{2}}\log z-{\frac {3z^{2}}{4}}+{\frac {z}{2}}\log 2\pi -{\frac {1}{12}}\log z\\&{}+\left({\frac {1}{12}}-\log A\right)+\sum _{k=1}^{N}{\frac {B_{2k+2}}{4k\left(k+1\right)z^{2k}}}~+~O\left({\frac {1}{z^{2N+2}}}\right).\end{aligned}}}

Here the B k {\displaystyle B_{k}} are the Bernoulli numbers and A {\displaystyle A} is the Glaisher–Kinkelin constant. (Note that somewhat confusingly at the time of Barnes [7] the Bernoulli number B 2 k {\displaystyle B_{2k}} would have been written as ( 1 ) k + 1 B k {\displaystyle (-1)^{k+1}B_{k}} , but this convention is no longer current.) This expansion is valid for z {\displaystyle z} in any sector not containing the negative real axis with | z | {\displaystyle |z|} large.

Relation to the log-gamma integral

The parametric log-gamma can be evaluated in terms of the Barnes G-function:[5]

0 z log Γ ( x ) d x = z ( 1 z ) 2 + z 2 log 2 π + z log Γ ( z ) log G ( 1 + z ) {\displaystyle \int _{0}^{z}\log \Gamma (x)\,dx={\frac {z(1-z)}{2}}+{\frac {z}{2}}\log 2\pi +z\log \Gamma (z)-\log G(1+z)}

The proof is somewhat indirect, and involves first considering the logarithmic difference of the gamma function and Barnes G-function:

z log Γ ( z ) log G ( 1 + z ) {\displaystyle z\log \Gamma (z)-\log G(1+z)}

where

1 Γ ( z ) = z e γ z k = 1 { ( 1 + z k ) e z / k } {\displaystyle {\frac {1}{\Gamma (z)}}=ze^{\gamma z}\prod _{k=1}^{\infty }\left\{\left(1+{\frac {z}{k}}\right)e^{-z/k}\right\}}

and γ {\displaystyle \,\gamma } is the Euler–Mascheroni constant.

Taking the logarithm of the Weierstrass product forms of the Barnes G-function and gamma function gives:

z log Γ ( z ) log G ( 1 + z ) = z log ( 1 Γ ( z ) ) log G ( 1 + z ) = z [ log z + γ z + k = 1 { log ( 1 + z k ) z k } ] [ z 2 log 2 π z 2 z 2 2 z 2 γ 2 + k = 1 { k log ( 1 + z k ) + z 2 2 k z } ] {\displaystyle {\begin{aligned}&z\log \Gamma (z)-\log G(1+z)=-z\log \left({\frac {1}{\Gamma (z)}}\right)-\log G(1+z)\\[5pt]={}&{-z}\left[\log z+\gamma z+\sum _{k=1}^{\infty }{\Bigg \{}\log \left(1+{\frac {z}{k}}\right)-{\frac {z}{k}}{\Bigg \}}\right]\\[5pt]&{}-\left[{\frac {z}{2}}\log 2\pi -{\frac {z}{2}}-{\frac {z^{2}}{2}}-{\frac {z^{2}\gamma }{2}}+\sum _{k=1}^{\infty }{\Bigg \{}k\log \left(1+{\frac {z}{k}}\right)+{\frac {z^{2}}{2k}}-z{\Bigg \}}\right]\end{aligned}}}

A little simplification and re-ordering of terms gives the series expansion:

k = 1 { ( k + z ) log ( 1 + z k ) z 2 2 k z } = z log z z 2 log 2 π + z 2 + z 2 2 z 2 γ 2 z log Γ ( z ) + log G ( 1 + z ) {\displaystyle {\begin{aligned}&\sum _{k=1}^{\infty }{\Bigg \{}(k+z)\log \left(1+{\frac {z}{k}}\right)-{\frac {z^{2}}{2k}}-z{\Bigg \}}\\[5pt]={}&{-z}\log z-{\frac {z}{2}}\log 2\pi +{\frac {z}{2}}+{\frac {z^{2}}{2}}-{\frac {z^{2}\gamma }{2}}-z\log \Gamma (z)+\log G(1+z)\end{aligned}}}

Finally, take the logarithm of the Weierstrass product form of the gamma function, and integrate over the interval [ 0 , z ] {\displaystyle \,[0,\,z]} to obtain:

0 z log Γ ( x ) d x = 0 z log ( 1 Γ ( x ) ) d x = ( z log z z ) z 2 γ 2 k = 1 { ( k + z ) log ( 1 + z k ) z 2 2 k z } {\displaystyle {\begin{aligned}&\int _{0}^{z}\log \Gamma (x)\,dx=-\int _{0}^{z}\log \left({\frac {1}{\Gamma (x)}}\right)\,dx\\[5pt]={}&{-(z\log z-z)}-{\frac {z^{2}\gamma }{2}}-\sum _{k=1}^{\infty }{\Bigg \{}(k+z)\log \left(1+{\frac {z}{k}}\right)-{\frac {z^{2}}{2k}}-z{\Bigg \}}\end{aligned}}}

Equating the two evaluations completes the proof:

0 z log Γ ( x ) d x = z ( 1 z ) 2 + z 2 log 2 π + z log Γ ( z ) log G ( 1 + z ) {\displaystyle \int _{0}^{z}\log \Gamma (x)\,dx={\frac {z(1-z)}{2}}+{\frac {z}{2}}\log 2\pi +z\log \Gamma (z)-\log G(1+z)}

And since G ( 1 + z ) = Γ ( z ) G ( z ) {\displaystyle \,G(1+z)=\Gamma (z)\,G(z)} then,

0 z log Γ ( x ) d x = z ( 1 z ) 2 + z 2 log 2 π ( 1 z ) log Γ ( z ) log G ( z ) . {\displaystyle \int _{0}^{z}\log \Gamma (x)\,dx={\frac {z(1-z)}{2}}+{\frac {z}{2}}\log 2\pi -(1-z)\log \Gamma (z)-\log G(z)\,.}

References

  1. ^ E. W. Barnes, "The theory of the G-function", Quarterly Journ. Pure and Appl. Math. 31 (1900), 264–314.
  2. ^ M. F. Vignéras, L'équation fonctionelle de la fonction zêta de Selberg du groupe mudulaire SL ( 2 , Z ) {\displaystyle (2,\mathbb {Z} )} , Astérisque 61, 235–249 (1979).
  3. ^ Park, Junesang (1996). "A duplication formula for the double gamma function $Gamma_2$". Bulletin of the Korean Mathematical Society. 33 (2): 289–294.
  4. ^ Marichal, Jean Luc. A Generalization of Bohr-Mollerup’s Theorem for Higher Order Convex Functions (PDF). Springer. p. 218.
  5. ^ a b Adamchik, Viktor S. (2003). "Contributions to the Theory of the Barnes function". arXiv:math/0308086.
  6. ^ I. Vardi, Determinants of Laplacians and multiple gamma functions, SIAM J. Math. Anal. 19, 493–507 (1988).
  7. ^ E. T. Whittaker and G. N. Watson, "A Course of Modern Analysis", CUP.