Introduction
In this paper we prove Nakamura’s conjecture, concerning the relation between the 2D Toda model and Yamazaki-Hori solutions of the Ernst equation. Recall that the Ernst equation is formulated in terms of a complex potential ξ \xi ξ , which allows reconstruction of metric functions for stationary axisymmetric gravitational fields. It has the form [2 ] :
( ξ ξ ∗ − 1 ) ∇ 2 ξ = 2 ξ ∗ ∇ ⃗ ξ ⋅ ∇ ⃗ ξ . (1) (\xi\xi^{*} - 1) \nabla^{2}\xi= 2\xi^{*}\vec{\nabla}\xi\cdot\vec{\nabla}\xi. \tag*{(1)} ( ξ ξ ∗ − 1 ) ∇ 2 ξ = 2 ξ ∗ ∇ ξ ⋅ ∇ ξ . (1)
The Yamazaki-Hori solutions to (1) are rational solutions parametrized by n ∈ Z ≥ 0 n \in\mathbb{Z}_{\geq0} n ∈ Z ≥ 0 , and have the form [3 , 4 ] :
ξ n = g n ( x , y , p , q ) f n ( x , y , p , q ) (2) \xi_n = \frac{g_n(x,y,p,q)}{f_n(x,y,p,q)} \tag*{(2)} ξ n = f n ( x , y , p , q ) g n ( x , y , p , q ) (2)
where x , y x, y x , y are the prolate spheroidal coordinates and p , q p, q p , q are two parameters satisfying p 2 + q 2 = 1 p^2+q^2=1 p 2 + q 2 = 1 .1 The semi-infinite 2D Toda molecule is described in terms of the τ \tau τ -functions τ n = τ n ( η , ζ ) \tau_n=\tau_n(\eta,\zeta) τ n = τ n ( η , ζ ) by the Hirota bilinear equation [5 ] :
D η D ζ τ n ⋅ τ n = 2 τ n + 1 τ n − 1 , τ 0 = 1. n = 1 , 2 , … , (3) D_\eta D_\zeta\tau_n\cdot\tau_n=2\tau_{n+1}\tau_{n-1},\qquad\tau_0=1.\qquad n=1,2,\ldots, \tag*{(3)} D η D ζ τ n ⋅ τ n = 2 τ n + 1 τ n − 1 , τ 0 = 1. n = 1 , 2 , … , (3)
with the solutions given by the following two-directional Wronskian:
τ n ( Φ ) = det T ( n ) ( Φ ) , T k l ( n ) ( Φ ) : = ( ∂ ∂ η ) k − 1 ( ∂ ∂ ζ ) l − 1 Φ ( η , ζ ) , 1 ≤ k , l ≤ n . (4) \tau_n(\Phi)=\det T^{(n)}(\Phi),\qquad T_{kl}^{(n)}(\Phi):=\left(\frac{\partial}{\partial\eta}\right)^{k-1}\left(\frac{\partial}{\partial\zeta}\right)^{l-1}\Phi(\eta,\zeta),\qquad1\leq k,l\leq n. \tag*{(4)} τ n ( Φ ) = det T ( n ) ( Φ ) , T k l ( n ) ( Φ ) := ( ∂ η ∂ ) k − 1 ( ∂ ζ ∂ ) l − 1 Φ ( η , ζ ) , 1 ≤ k , l ≤ n . (4)
Here Φ ( η , ζ ) \Phi(\eta,\zeta) Φ ( η , ζ ) is an arbitrary seed function.2
Nakamura conjectured and verified numerically for the first values of n n n in [5 ] that the polynomials g n g_n g n and f n f_n f n can be obtained from the τ \tau τ -function of the 2D Toda model, thus establishing a relation between the two models. Fukuyama, Kamimura and Yu [6 ] proved the conjecture for the nonrotating Weyl case q = 0 q=0 q = 0 . Subsequently, Fukuyama and Koizumi [7 ] established the conjectured relations for the highest and lowest orders in an expansion in the rotation parameter. More recently, Fukuyama [8 ] investigated the underlying determinant structure and its relation to the conjecture.
In this paper we prove the conjecture for arbitrary values of the parameter s = q / p s=q/p s = q / p ,3 using the Pfaffian solutions to the Ernst equation obtained in [1 ] . We recall the main result of [1 ] in the following theorem.
Theorem 1 (Pfaffian form of the Yamazaki-Hori solutions). The Yamazaki-Hori solutions are expressed in terms of the polynomials f n f_n f n and g n g_n g n , each given by a product of Pfaffians, while the corresponding Ernst potential ξ n = g n / f n \xi_n=g_n/f_n ξ n = g n / f n is given by the ratio of two Pfaffians. In particular, we have the following two cases.
Even case (m = n − 1 = 2 k m=n-1=2k m = n − 1 = 2 k ):
f n = C ( f ) pf ‾ ( ω ( s ) ) pf ( ω ( s ) ) , C ( f ) = ( − 1 ) n p 2 n − 2 C ( n ) 2 n − 1 , (5) f_n=C^{(f)}\overline{\operatorname{pf}}(\omega^{(s)})\operatorname{pf}(\omega^{(s)}),\qquad C^{(f)}=\frac{(-1)^n p^{2n-2}C(n)}{2n-1}, \tag*{(5)} f n = C ( f ) pf ( ω ( s ) ) pf ( ω ( s ) ) , C ( f ) = 2 n − 1 ( − 1 ) n p 2 n − 2 C ( n ) , (5) g n = C ( g ) pf ‾ ( ω ( s ) ) pf ( Ω ( e ) ) , C ( g ) = ( − 1 ) n p 2 n − 1 C ( n ) . (6) g_n=C^{(g)}\overline{\operatorname{pf}}(\omega^{(s)})\operatorname{pf}(\Omega^{(e)}),\qquad C^{(g)}=(-1)^n p^{2n-1}C(n). \tag*{(6)} g n = C ( g ) pf ( ω ( s ) ) pf ( Ω ( e ) ) , C ( g ) = ( − 1 ) n p 2 n − 1 C ( n ) . (6) Thus, for the Ernst potential:
ξ n = p ( 2 n − 1 ) pf ( Ω ( e ) ) pf ( ω ( s ) ) . (7) \xi_n=p(2n-1)\frac{\operatorname{pf}(\Omega^{(e)})}{\operatorname{pf}(\omega^{(s)})}. \tag*{(7)} ξ n = p ( 2 n − 1 ) pf ( ω ( s ) ) pf ( Ω ( e ) ) . (7) Odd case (m = n − 1 = 2 k + 1 m=n-1=2k+1 m = n − 1 = 2 k + 1 ):
f n = C ~ ( f ) pf ‾ ( ω ^ ( s ) ) pf ( ω ^ ( s ) ) , C ~ ( f ) = ( − 1 ) n p 2 n − 2 C ( n ) 2 ( 2 n − 1 ) , (8) f_n=\widetilde{C}^{(f)}\overline{\operatorname{pf}}(\widehat{\omega}^{(s)})\operatorname{pf}(\widehat{\omega}^{(s)}),\qquad\widetilde{C}^{(f)}=\frac{(-1)^n p^{2n-2}C(n)}{2(2n-1)}, \tag*{(8)} f n = C ( f ) pf ( ω ( s ) ) pf ( ω ( s ) ) , C ( f ) = 2 ( 2 n − 1 ) ( − 1 ) n p 2 n − 2 C ( n ) , (8) g n = C ~ ( g ) pf ‾ ( ω ^ ( s ) ) pf ( Ω ( o ) ) , C ~ ( g ) = ( − 1 ) n p 2 n − 1 C ( n ) . (9) g_n=\widetilde{C}^{(g)}\overline{\operatorname{pf}}(\widehat{\omega}^{(s)})\operatorname{pf}(\Omega^{(o)}),\qquad\widetilde{C}^{(g)}=(-1)^n p^{2n-1}C(n). \tag*{(9)} g n = C ( g ) pf ( ω ( s ) ) pf ( Ω ( o ) ) , C ( g ) = ( − 1 ) n p 2 n − 1 C ( n ) . (9)
Thus, for the Ernst potential:
ξ n = 2 p ( 2 n − 1 ) pf ( Ω ( o ) ) pf ( ω ^ ( s ) ) . (10) \xi_n = 2p(2n-1)\frac{\operatorname{pf}\left(\Omega^{(o)}\right)}{\operatorname{pf}\left(\widehat{\omega}^{(s)}\right)}. \tag*{(10)} ξ n = 2 p ( 2 n − 1 ) pf ( ω ( s ) ) pf ( Ω ( o ) ) . (10)
In these formulas, the constant C ( n ) C(n) C ( n ) is given by the following expression:
C ( n ) : = ∏ j = 1 n − 1 [ − 2 ( 2 j − 1 ) ] = ( − 1 ) n − 1 Γ ( 2 n − 1 ) Γ ( n ) , (11) C(n) := \prod_{j=1}^{n-1}\left[-2(2j-1)\right] = (-1)^{n-1}\frac{\Gamma(2n-1)}{\Gamma(n)}, \tag*{(11)} C ( n ) := j = 1 ∏ n − 1 [ − 2 ( 2 j − 1 ) ] = ( − 1 ) n − 1 Γ ( n ) Γ ( 2 n − 1 ) , (11)
and the explicit forms of the Pfaffians in the above formulas are collected in Appendix A .
We use the above theorem in order to prove Nakamura’s conjecture, which we formulate as follows:
Theorem 2 (Nakamura’s conjecture). The Yamazaki-Hori Ernst potential is related to the τ \tau τ -functions of the 2D Toda model by:
ξ n = − τ n ( Φ ) τ n − 1 ( Φ ′ ) , n ≥ 1. (12) \xi_n = -\frac{\tau_n(\Phi)}{\tau_{n-1}(\Phi')}, \qquad n \ge1. \tag*{(12)} ξ n = − τ n − 1 ( Φ ′ ) τ n ( Φ ) , n ≥ 1. (12) More precisely, setting m = n − 1 m=n-1 m = n − 1 , the following identities hold:
Even case (m = 2 k m=2k m = 2 k ):
τ m ( Φ ′ ) = N m pf ( ω ( s ) ) , (13) \tau_m(\Phi') = \mathcal{N}_m \operatorname{pf}\left(\omega^{(s)}\right), \tag*{(13)} τ m ( Φ ′ ) = N m pf ( ω ( s ) ) , (13) τ n ( Φ ) = − p ( 2 m + 1 ) N m pf ( Ω ( e ) ) . (14) \tau_n(\Phi) = -p(2m+1)\mathcal{N}_m \operatorname{pf}\left(\Omega^{(e)}\right). \tag*{(14)} τ n ( Φ ) = − p ( 2 m + 1 ) N m pf ( Ω ( e ) ) . (14) Odd case (m = 2 k + 1 m=2k+1 m = 2 k + 1 ):
τ m ( Φ ′ ) = N m pf ( ω ^ ( s ) ) , (15) \tau_m(\Phi') = \mathcal{N}_m \operatorname{pf}\left(\widehat{\omega}^{(s)}\right), \tag*{(15)} τ m ( Φ ′ ) = N m pf ( ω ( s ) ) , (15) τ n ( Φ ) = − 2 p ( 2 m + 1 ) N m pf ( Ω ( o ) ) . (16) \tau_n(\Phi) = -2p(2m+1)\mathcal{N}_m \operatorname{pf}\left(\Omega^{(o)}\right). \tag*{(16)} τ n ( Φ ) = − 2 p ( 2 m + 1 ) N m pf ( Ω ( o ) ) . (16) Here the normalization factor N m \mathcal{N}_m N m is given by
N m = 2 ⌊ m / 2 ⌋ p m ∏ j = 1 m ( 2 j − 1 ) ( 2 j ) ! , N 0 = 1 , (17) \mathcal{N}_m = 2^{\lfloor m/2\rfloor}p^m\prod_{j=1}^{m}(2j-1)(2j)!, \qquad\mathcal{N}_0=1, \tag*{(17)} N m = 2 ⌊ m /2 ⌋ p m j = 1 ∏ m ( 2 j − 1 ) ( 2 j )! , N 0 = 1 , (17) where ⌊ m / 2 ⌋ \lfloor m/2\rfloor ⌊ m /2 ⌋ denotes the integer part of m / 2 m/2 m /2 . The two seed functions Φ \Phi Φ and Φ ′ = ∂ ζ ∂ η Φ \Phi'=\partial_\zeta\partial_\eta\Phi Φ ′ = ∂ ζ ∂ η Φ are given by:
Φ ( x , y ) = − p x + i q y , (18) \Phi(x,y)=-px+iqy, \tag*{(18)} Φ ( x , y ) = − p x + i q y , (18) Φ ′ ( x , y ) = − 2 p x ( x 2 − 1 ) − 2 i q y ( y 2 − 1 ) . (19) \Phi'(x,y)=-2px(x^2-1)-2iqy(y^2-1). \tag*{(19)} Φ ′ ( x , y ) = − 2 p x ( x 2 − 1 ) − 2 i q y ( y 2 − 1 ) . (19)
Proof of Nakamura's conjecture
Polynomial identities and the Toda matrix
First, we consider the even case corresponding to m = n − 1 = 2 k m=n-1=2k m = n − 1 = 2 k , and the polynomial f n f_n f n . According to Theorem 1:
f n = C ( f ) pf ( ω ( s ) ) ‾ pf ( ω ( s ) ) , C ( f ) = ( − 1 ) n p 2 n − 2 C ( n ) 2 n − 1 , (20) f_n=C^{(f)}\overline{\operatorname{pf}\left(\omega^{(s)}\right)}\operatorname{pf}\left(\omega^{(s)}\right), \qquad C^{(f)}=\frac{(-1)^n p^{2n-2}C(n)}{2n-1}, \tag*{(20)} f n = C ( f ) pf ( ω ( s ) ) pf ( ω ( s ) ) , C ( f ) = 2 n − 1 ( − 1 ) n p 2 n − 2 C ( n ) , (20)
where (see Appendix A for details):
ω i j ( s ) = ω i j − i s 2 [ Γ i ( x ) Γ j ( y ) − Γ i ( y ) Γ j ( x ) ] , Γ k ( w ) : = 1 2 k − 1 [ w 2 k − 1 − w 2 k + 1 ] (21) \omega_{ij}^{(s)}=\omega_{ij}-\frac{\mathrm{i}s}{2}\left[\Gamma_i(x)\Gamma_j(y)-\Gamma_i(y)\Gamma_j(x)\right],\qquad\Gamma_k(w):=\frac{1}{2k-1}\left[w^{2k-1}-w^{2k+1}\right] \tag*{(21)} ω ij ( s ) = ω ij − 2 i s [ Γ i ( x ) Γ j ( y ) − Γ i ( y ) Γ j ( x ) ] , Γ k ( w ) := 2 k − 1 1 [ w 2 k − 1 − w 2 k + 1 ] (21)
ω i j = 1 2 I i j ( x ) + s 2 2 I i j ( y ) , I i j ( z ) : = ∫ 1 z [ Γ i ( w ) Γ j ′ ( w ) − Γ i ′ ( w ) Γ j ( w ) ] d w (22) \omega_{ij}=\frac{1}{2}I_{ij}(x)+\frac{s^2}{2}I_{ij}(y),\qquad I_{ij}(z):=\int_1^z\left[\Gamma_i(w)\Gamma_j'(w)-\Gamma_i'(w)\Gamma_j(w)\right]\,\mathrm{d}w \tag*{(22)} ω ij = 2 1 I ij ( x ) + 2 s 2 I ij ( y ) , I ij ( z ) := ∫ 1 z [ Γ i ( w ) Γ j ′ ( w ) − Γ i ′ ( w ) Γ j ( w ) ] d w (22)
We now relate the polynomials Γ k ( z ) \Gamma_k(z) Γ k ( z ) to the polynomials P n ( z ) P_n(z) P n ( z ) , which arise in the Toda τ \tau τ -function associated with the seed Φ ′ \Phi' Φ ′ and are defined by the following recurrence relation.4
P k + 1 ( z ) = ( z 2 − 1 ) d d z P k ( z ) , P 0 ( z ) = z , (23) P_{k+1}(z)=(z^2-1)\frac{\mathrm{d}}{\mathrm{d}z}P_k(z),\qquad P_0(z)=z, \tag*{(23)} P k + 1 ( z ) = ( z 2 − 1 ) d z d P k ( z ) , P 0 ( z ) = z , (23)
For the seed function Φ ′ \Phi' Φ ′ (19) , one finds:5
∂ η a ∂ ζ b Φ ′ = − p [ P a + b + 2 ( x ) + i s ( − 1 ) b P a + b + 2 ( y ) ] . (24) \partial_\eta^a\partial_\zeta^b\Phi'=-p\left[P_{a+b+2}(x)+\mathrm{i}s(-1)^bP_{a+b+2}(y)\right]. \tag*{(24)} ∂ η a ∂ ζ b Φ ′ = − p [ P a + b + 2 ( x ) + i s ( − 1 ) b P a + b + 2 ( y ) ] . (24)
Denoting:
U j = P j + 2 ( x ) + i s P j + 2 ( y ) , (25) U_j=P_{j+2}(x)+\mathrm{i}sP_{j+2}(y), \tag*{(25)} U j = P j + 2 ( x ) + i s P j + 2 ( y ) , (25)
V j = P j + 2 ( x ) − i s P j + 2 ( y ) . (26) V_j=P_{j+2}(x)-\mathrm{i}sP_{j+2}(y). \tag*{(26)} V j = P j + 2 ( x ) − i s P j + 2 ( y ) . (26)
the τ \tau τ -function of the 2D Toda model can be written in the form:
τ 2 k ( Φ ′ ) = ( − p ) 2 k det T 2 k ( Φ ′ ) , (27) \tau_{2k}(\Phi')=(-p)^{2k}\det\mathcal{T}_{2k}^{(\Phi')}, \tag*{(27)} τ 2 k ( Φ ′ ) = ( − p ) 2 k det T 2 k ( Φ ′ ) , (27)
where the 2 k × 2 k 2k\times2k 2 k × 2 k Toda matrix T 2 k ( Φ ′ ) \mathcal{T}_{2k}^{(\Phi')} T 2 k ( Φ ′ ) has the form:
T 2 k ( Φ ′ ) = ( U 0 V 1 U 2 V 3 ⋯ U m − 2 V m − 1 U 1 V 2 U 3 V 4 ⋯ U m − 1 V m U 2 V 3 U 4 V 5 ⋯ U m V m + 1 ⋮ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮ U m − 2 V m − 1 U m V m + 1 ⋯ U 2 m − 4 V 2 m − 3 U m − 1 V m U m + 1 V m + 2 ⋯ U 2 m − 3 V 2 m − 2 ) , m = 2 k . (28) \mathcal{T}_{2k}^{(\Phi')}=
\begin{pmatrix}
U_0 & V_1 & U_2 & V_3 & \cdots& U_{m-2} & V_{m-1}\\
U_1 & V_2 & U_3 & V_4 & \cdots& U_{m-1} & V_m\\
U_2 & V_3 & U_4 & V_5 & \cdots& U_m & V_{m+1}\\
\vdots& \vdots& \vdots& \vdots& \ddots& \vdots& \vdots\\
U_{m-2} & V_{m-1} & U_m & V_{m+1} & \cdots& U_{2m-4} & V_{2m-3}\\
U_{m-1} & V_m & U_{m+1} & V_{m+2} & \cdots& U_{2m-3} & V_{2m-2}
\end{pmatrix},\qquad m=2k. \tag*{(28)} T 2 k ( Φ ′ ) = U 0 U 1 U 2 ⋮ U m − 2 U m − 1 V 1 V 2 V 3 ⋮ V m − 1 V m U 2 U 3 U 4 ⋮ U m U m + 1 V 3 V 4 V 5 ⋮ V m + 1 V m + 2 ⋯ ⋯ ⋯ ⋱ ⋯ ⋯ U m − 2 U m − 1 U m ⋮ U 2 m − 4 U 2 m − 3 V m − 1 V m V m + 1 ⋮ V 2 m − 3 V 2 m − 2 , m = 2 k . (28)
Noting that deg P j = j + 1 \deg P_j=j+1 deg P j = j + 1 and deg Γ j = 2 j + 1 \deg\Gamma_j=2j+1 deg Γ j = 2 j + 1 , and writing P 2 j − 1 ( z ) = ∑ ℓ = 0 j a 2 ℓ ( j ) z 2 ℓ P_{2j-1}(z)=\sum_{\ell=0}^{j}a_{2\ell}^{(j)}z^{2\ell} P 2 j − 1 ( z ) = ∑ ℓ = 0 j a 2 ℓ ( j ) z 2 ℓ , for j = 1 , … , m j=1,\ldots,m j = 1 , … , m , it follows directly from the recurrence relation (23) that the two polynomial sequences are related by:
P 2 j ( z ) = ∑ ℓ = 1 j Λ j ℓ ( m ) Γ ℓ ( z ) , j = 1 , … , m , (29) P_{2j}(z)=\sum_{\ell=1}^{j}\Lambda_{j\ell}^{(m)}\Gamma_\ell(z),\qquad j=1,\ldots,m, \tag*{(29)} P 2 j ( z ) = ℓ = 1 ∑ j Λ j ℓ ( m ) Γ ℓ ( z ) , j = 1 , … , m , (29)
Λ j ℓ ( m ) = − 2 ℓ ( 2 ℓ − 1 ) a 2 ℓ ( j ) , 1 ≤ ℓ ≤ j ≤ m . (30) \Lambda_{j\ell}^{(m)}=-2\ell(2\ell-1)a_{2\ell}^{(j)},\qquad1\leq\ell\leq j\leq m. \tag*{(30)} Λ j ℓ ( m ) = − 2 ℓ ( 2 ℓ − 1 ) a 2 ℓ ( j ) , 1 ≤ ℓ ≤ j ≤ m . (30)
∂ η = ( x 2 − 1 ) ∂ x + ( y 2 − 1 ) ∂ y , \partial_\eta=(x^2-1)\partial_x+(y^2-1)\partial_y, ∂ η = ( x 2 − 1 ) ∂ x + ( y 2 − 1 ) ∂ y ,
∂ ζ = ( x 2 − 1 ) ∂ x − ( y 2 − 1 ) ∂ y . \partial_\zeta=(x^2-1)\partial_x-(y^2-1)\partial_y. ∂ ζ = ( x 2 − 1 ) ∂ x − ( y 2 − 1 ) ∂ y .
Thus, Λ ( m ) \Lambda^{(m)} Λ ( m ) is a lower-triangular matrix. Since a 2 j ( j ) = ( 2 j − 1 ) ! a_{2j}^{(j)}=(2j-1)! a 2 j ( j ) = ( 2 j − 1 )! , we find:
det Λ ( m ) = ( − 1 ) m ∏ ℓ = 1 m ( 2 ℓ ) ! ( 2 ℓ − 1 ) (31) \det\Lambda^{(m)}=(-1)^m\prod_{\ell=1}^{m}(2\ell)!(2\ell-1) \tag*{(31)} det Λ ( m ) = ( − 1 ) m ℓ = 1 ∏ m ( 2 ℓ )! ( 2 ℓ − 1 ) (31)
With the explicit formulas relating the polynomial sequences P n P_n P n and Γ n \Gamma_n Γ n , we can evaluate ω ( s ) \omega^{(s)} ω ( s ) in terms of the polynomials P n P_n P n , which is the first step to compare pf ( ω ( s ) ) \operatorname{pf}(\omega^{(s)}) pf ( ω ( s ) ) with the determinant of the matrix T 2 k ( Φ ′ ) T_{2k}^{(\Phi') } T 2 k ( Φ ′ ) . Defining (cf (21) ):
I i j ( P ) ( z ) : = ∫ 1 z [ P i ( w ) P j ′ ( w ) − P i ′ ( w ) P j ( w ) ] d w (32) \mathcal{I}_{ij}^{(P)}(z):=\int_{1}^{z}\left[P_i(w)P_j'(w)-P_i'(w)P_j(w)\right]\,\mathrm{d}w \tag*{(32)} I ij ( P ) ( z ) := ∫ 1 z [ P i ( w ) P j ′ ( w ) − P i ′ ( w ) P j ( w ) ] d w (32)
and using the relation between P 2 i P_{2i} P 2 i and Γ j \Gamma_j Γ j (29), we find:
I 2 r , 2 s ( P ) ( z ) = ∑ i , j = 1 m Λ r i ( m ) Λ s j ( m ) I i j ( z ) , 1 ≤ r , s ≤ m . (33) \mathcal{I}_{2r,2s}^{(P)}(z)=\sum_{i,j=1}^{m}\Lambda_{ri}^{(m)}\Lambda_{sj}^{(m)}\mathcal{I}_{ij}(z),\qquad1\le r,s\le m. \tag*{(33)} I 2 r , 2 s ( P ) ( z ) = i , j = 1 ∑ m Λ r i ( m ) Λ s j ( m ) I ij ( z ) , 1 ≤ r , s ≤ m . (33)
Integration by parts and the recurrence relation (23) give:
I r s ( P ) ( z ) = P r ( z ) P s ( z ) − P r + 1 ( z ) P s − 1 ( z ) − I r + 1 , s − 1 ( P ) ( z ) . (34) \mathcal{I}_{rs}^{(P)}(z)=P_r(z)P_s(z)-P_{r+1}(z)P_{s-1}(z)-\mathcal{I}_{r+1,s-1}^{(P)}(z). \tag*{(34)} I r s ( P ) ( z ) = P r ( z ) P s ( z ) − P r + 1 ( z ) P s − 1 ( z ) − I r + 1 , s − 1 ( P ) ( z ) . (34)
Applying repeatedly the above formula, and noting that for even s − r s-r s − r , the iteration terminates with I ( r + s ) / 2 , ( r + s ) / 2 ( P ) ( z ) = 0 \mathcal{I}_{(r+s)/2,(r+s)/2}^{(P)}(z)=0 I ( r + s ) /2 , ( r + s ) /2 ( P ) ( z ) = 0 , one obtains (note that the products P r + j P s − j P_{r+j}P_{s-j} P r + j P s − j with 1 ≤ j < ( s − r ) / 2 1\le j<(s-r)/2 1 ≤ j < ( s − r ) /2 occur twice, while P r P s P_rP_s P r P s and P ( r + s ) / 2 2 P_{(r+s)/2}^{2} P ( r + s ) /2 2 occur only once):
I r s ( P ) ( z ) = ∑ j = 0 s − r − 1 ( − 1 ) j P r + j ( z ) P s − j ( z ) , 0 ≤ r < s , s − r even . (35) \mathcal{I}_{rs}^{(P)}(z)=\sum_{j=0}^{s-r-1}(-1)^jP_{r+j}(z)P_{s-j}(z),\qquad0\le r<s,\quad s-r\ \text{even}. \tag*{(35)} I r s ( P ) ( z ) = j = 0 ∑ s − r − 1 ( − 1 ) j P r + j ( z ) P s − j ( z ) , 0 ≤ r < s , s − r even . (35)
Using the above formula, and the definitions of U j U_j U j and V j V_j V j , we find:
2 ∑ i , j = 1 m Λ a i ( m ) Λ b j ( m ) ω i j ( s ) = ∑ ℓ = 0 2 ( b − a ) − 1 ( − 1 ) ℓ U 2 a − 2 + ℓ V 2 b − 2 − ℓ , 1 ≤ a < b ≤ m . (36) 2\sum_{i,j=1}^{m}\Lambda_{ai}^{(m)}\Lambda_{bj}^{(m)}\omega_{ij}^{(s)}
=\sum_{\ell=0}^{2(b-a)-1}(-1)^\ell U_{2a-2+\ell}V_{2b-2-\ell},\qquad1\le a<b\le m. \tag*{(36)} 2 i , j = 1 ∑ m Λ ai ( m ) Λ bj ( m ) ω ij ( s ) = ℓ = 0 ∑ 2 ( b − a ) − 1 ( − 1 ) ℓ U 2 a − 2 + ℓ V 2 b − 2 − ℓ , 1 ≤ a < b ≤ m . (36)
Extended Cusick identities
The following generalized Cusick identity was established in [9 ] :
Theorem 3 (Generalized Cusick identity). Let ϕ r , ψ r \phi_r,\psi_r ϕ r , ψ r , r ≥ 1 r\ge1 r ≥ 1 be arbitrary sequences, and C ( ϕ , ψ ) = ( c a b ) C(\phi,\psi)=(c_{ab}) C ( ϕ , ψ ) = ( c ab ) the antisymmetric matrix of order 2 k 2k 2 k , with entries:
c a b = ∑ ℓ = 1 b − a ϕ a + ℓ − 1 ψ b − ℓ , 1 ≤ a < b ≤ 2 k . (37) c_{ab}=\sum_{\ell=1}^{b-a}\phi_{a+\ell-1}\psi_{b-\ell},\qquad1\le a<b\le2k. \tag*{(37)} c ab = ℓ = 1 ∑ b − a ϕ a + ℓ − 1 ψ b − ℓ , 1 ≤ a < b ≤ 2 k . (37) The generalized Cusick identity states that:
pf C ( ϕ , ψ ) = det ( ϕ i + j + 1 ) i , j = 0 k − 1 det ( ψ i + j + 1 ) i , j = 0 k − 1 . (38) \operatorname{pf} C(\phi,\psi)=\det\left(\phi_{i+j+1}\right)_{i,j=0}^{k-1}\det\left(\psi_{i+j+1}\right)_{i,j=0}^{k-1}. \tag*{(38)} pf C ( ϕ , ψ ) = det ( ϕ i + j + 1 ) i , j = 0 k − 1 det ( ψ i + j + 1 ) i , j = 0 k − 1 . (38)
Setting u j = ϕ j + 1 u_j=\phi_{j+1} u j = ϕ j + 1 , v j = ψ j + 1 v_j=\psi_{j+1} v j = ψ j + 1 , and shifting the summation index ℓ → ℓ + 1 \ell\to\ell+1 ℓ → ℓ + 1 , we define the antisymmetric matrix C ( u , v ) C(u,v) C ( u , v ) of arbitrary order m m m by
C ( u , v ) a b : = ∑ ℓ = 0 b − a − 1 u a − 1 + ℓ v b − 2 − ℓ , 1 ≤ a < b ≤ m , (39) C(u,v)_{ab}:=\sum_{\ell=0}^{b-a-1}u_{a-1+\ell}v_{b-2-\ell},\qquad1\le a<b\le m, \tag*{(39)} C ( u , v ) ab := ℓ = 0 ∑ b − a − 1 u a − 1 + ℓ v b − 2 − ℓ , 1 ≤ a < b ≤ m , (39)
with antisymmetric extension to a > b a>b a > b . In this notation, the generalized Cusick identity becomes:
pf C ( u , v ) = det U det V , (40) \operatorname{pf}C(u,v)=\det\mathcal{U}\det\mathcal{V}, \tag*{(40)} pf C ( u , v ) = det U det V , (40)
where U i j = u i + j \mathcal{U}_{ij}=u_{i+j} U ij = u i + j and V i j = v i + j \mathcal{V}_{ij}=v_{i+j} V ij = v i + j , for 0 ≤ i , j < k 0\le i,j<k 0 ≤ i , j < k .
We now show that the sum in (36) can be expressed as a difference of two generalized Cusick sums. To this end, we introduce four sequences:
u j = U 2 j , p j = V 2 j + 1 , q j = U 2 j + 1 , v j = V 2 j + 2 . (41) u_j=U_{2j},\qquad p_j=V_{2j+1},\qquad q_j=U_{2j+1},\qquad v_j=V_{2j+2}. \tag*{(41)} u j = U 2 j , p j = V 2 j + 1 , q j = U 2 j + 1 , v j = V 2 j + 2 . (41)
Then, the equation (36) becomes:
2 ∑ i , j = 1 m Λ a i ( m ) Λ b j ( m ) ω i j ( s ) = ∑ ℓ = 0 b − a − 1 ( u a − 1 + ℓ v b − 2 − ℓ − q a − 1 + ℓ p b − 2 − ℓ ) , 1 ≤ a < b ≤ m . (42) 2\sum_{i,j=1}^{m}\Lambda_{ai}^{(m)}\Lambda_{bj}^{(m)}\omega_{ij}^{(s)}
=\sum_{\ell=0}^{b-a-1}\left(u_{a-1+\ell}v_{b-2-\ell}-q_{a-1+\ell}p_{b-2-\ell}\right),\qquad1\le a<b\le m. \tag*{(42)} 2 i , j = 1 ∑ m Λ ai ( m ) Λ bj ( m ) ω ij ( s ) = ℓ = 0 ∑ b − a − 1 ( u a − 1 + ℓ v b − 2 − ℓ − q a − 1 + ℓ p b − 2 − ℓ ) , 1 ≤ a < b ≤ m . (42)
Using the definition of C ( u , v ) C(u,v) C ( u , v ) , we write the formula (42) in the matrix form:6
2 Λ ( m ) ω ( s ) ( Λ ( m ) ) T = C ( u , v ) − C ( q , p ) . (43) 2\Lambda^{(m)}\omega^{(s)}\left(\Lambda^{(m)}\right)^{T}=C(u,v)-C(q,p). \tag*{(43)} 2 Λ ( m ) ω ( s ) ( Λ ( m ) ) T = C ( u , v ) − C ( q , p ) . (43)
This relation motivates the following extension of Dale’s generalized Cusick identity to four arbitrary sequences.
Theorem 4 (Extended Cusick identity). Let m = 2 k m=2k m = 2 k , let u j , p j , q j , v j u_j,p_j,q_j,v_j u j , p j , q j , v j , 0 ≤ j ≤ m − 2 0\le j\le m-2 0 ≤ j ≤ m − 2 , be four arbitrary sequences, and let C ( u , v ) C(u,v) C ( u , v ) and C ( q , p ) C(q,p) C ( q , p ) be antisymmetric matrices of order m m m , defined as above. Then:
pf [ C ( u , v ) − C ( q , p ) ] = det ( U P Q V ) , (44) \operatorname{pf}[C(u,v)-C(q,p)]=\det\begin{pmatrix}\mathcal{U}&\mathcal{P}\\\mathcal{Q}&\mathcal{V}\end{pmatrix}, \tag*{(44)} pf [ C ( u , v ) − C ( q , p )] = det ( U Q P V ) , (44) where U , P , Q , V \mathcal{U},\mathcal{P},\mathcal{Q},\mathcal{V} U , P , Q , V are k × k k\times k k × k Hankel matrices with entries
U i j = u i + j , P i j = p i + j , Q i j = q i + j , V i j = v i + j , 0 ≤ i , j < k . (45) \mathcal{U}_{ij}=u_{i+j},\qquad\mathcal{P}_{ij}=p_{i+j},\qquad\mathcal{Q}_{ij}=q_{i+j},\qquad\mathcal{V}_{ij}=v_{i+j},\qquad0\le i,j<k. \tag*{(45)} U ij = u i + j , P ij = p i + j , Q ij = q i + j , V ij = v i + j , 0 ≤ i , j < k . (45)
Setting p j = q j = 0 p_j=q_j=0 p j = q j = 0 , the determinant on the right-hand side becomes block diagonal, and we recover the generalized Cusick identity of [9 ] . The extended Cusick identity also implies a corresponding bordered identity for odd m = 2 k + 1 m=2k+1 m = 2 k + 1 .
Theorem 5 (Bordered extended Cusick identity). Let m = 2 k + 1 m=2k+1 m = 2 k + 1 , let u j u_j u j , 0 ≤ j ≤ 2 k 0\le j\le2k 0 ≤ j ≤ 2 k , and p j , q j , v j p_j,q_j,v_j p j , q j , v j , 0 ≤ j ≤ 2 k − 1 0\le j\le2k-1 0 ≤ j ≤ 2 k − 1 , be arbitrary sequences, and let C m = C ( u , v ) − C ( q , p ) C_m=C(u,v)-C(q,p) C m = C ( u , v ) − C ( q , p ) be the antisymmetric matrix of order m m m , defined as above. Introducing the column vector
u = ( u 0 , u 1 , … , u m − 1 ) T , (46) \mathbf{u}=(u_0,u_1,\ldots,u_{m-1})^{T}, \tag*{(46)} u = ( u 0 , u 1 , … , u m − 1 ) T , (46)
we have
pf ( C m u − u T 0 ) = det ( U P Q V ) (47) \operatorname{pf}\begin{pmatrix} C_m & \mathbf{u} \\ -\mathbf{u}^{\mathsf{T}} & 0 \end{pmatrix}=\det\begin{pmatrix} \mathcal{U} & \mathcal{P} \\ \mathcal{Q} & \mathcal{V} \end{pmatrix} \tag*{(47)} pf ( C m − u T u 0 ) = det ( U Q P V ) (47)
where the four Hankel blocks U , P , Q , V \mathcal{U},\mathcal{P},\mathcal{Q},\mathcal{V} U , P , Q , V have dimensions ( k + 1 ) × ( k + 1 ) (k+1)\times(k+1) ( k + 1 ) × ( k + 1 ) , ( k + 1 ) × k (k+1)\times k ( k + 1 ) × k , k × ( k + 1 ) k\times(k+1) k × ( k + 1 ) , and k × k k\times k k × k , respectively, with entries:
U = ( u i + j ) i , j = 0 k , P = ( p i + j ) 0 ≤ i ≤ k 0 ≤ j < k (48) \mathcal{U}=(u_{i+j})_{i,j=0}^{k},\qquad\mathcal{P}=(p_{i+j})_{\substack{0\leq i\leq k\\0\leq j<k}} \tag*{(48)} U = ( u i + j ) i , j = 0 k , P = ( p i + j ) 0 ≤ i ≤ k 0 ≤ j < k (48)
Q = ( q i + j ) 0 ≤ i < k 0 ≤ j ≤ k , V = ( v i + j ) i , j = 0 k − 1 . (49) \mathcal{Q}=(q_{i+j})_{\substack{0\leq i<k\\0\leq j\leq k}},\qquad\mathcal{V}=(v_{i+j})_{i,j=0}^{k-1}. \tag*{(49)} Q = ( q i + j ) 0 ≤ i < k 0 ≤ j ≤ k , V = ( v i + j ) i , j = 0 k − 1 . (49)
The proofs of both extended Cusick identities are given in Appendix B.
Nakamura’s conjecture follows from the two extended Cusick identities stated above. We consider each case corresponding to the seed functions Φ ′ \Phi' Φ ′ and Φ \Phi Φ separately.
Proof for the seed Φ ′ \Phi' Φ ′
Lemma 6. Let m = n − 1 m=n-1 m = n − 1 , and consider the Toda τ \tau τ -function associated with the seed Φ ′ \Phi' Φ ′ . Then.7
det T m ( Φ ′ ) = { 2 k det Λ ( m ) pf ( ω ( s ) ) , m = 2 k , 2 k det Λ ( m ) pf ( ω ^ ( s ) ) , m = 2 k + 1. (50) \det\mathcal{T}_{m}^{(\Phi')}=
\begin{cases}
2^k\det\Lambda^{(m)}\operatorname{pf}\left(\omega^{(s)}\right), & m=2k,\\
2^k\det\Lambda^{(m)}\operatorname{pf}\left(\widehat{\omega}^{(s)}\right), & m=2k+1.
\end{cases} \tag*{(50)} det T m ( Φ ′ ) = { 2 k det Λ ( m ) pf ( ω ( s ) ) , 2 k det Λ ( m ) pf ( ω ( s ) ) , m = 2 k , m = 2 k + 1. (50) where Λ ( m ) \Lambda^{(m)} Λ ( m ) is the lower-triangular matrix defined in (30) , and ω ( s ) \omega^{(s)} ω ( s ) , ω ^ ( s ) \widehat{\omega}^{(s)} ω ( s ) are antisymmetric matrices defined in Appendix A.
Proof.
Even case (m = 2 k m=2k m = 2 k ). By elementary row and column permutations, the Toda matrix T 2 k ( Φ ′ ) \mathcal{T}_{2k}^{(\Phi')} T 2 k ( Φ ′ ) (28) can be brought into the block-Hankel form:
det T 2 k ( Φ ′ ) = det ( U P Q V ) , (51) \det\mathcal{T}_{2k}^{(\Phi')}=\det\begin{pmatrix}\mathcal{U}&\mathcal{P}\\\mathcal{Q}&\mathcal{V}\end{pmatrix}, \tag*{(51)} det T 2 k ( Φ ′ ) = det ( U Q P V ) , (51)
where, for 0 ≤ i , j < k 0\leq i,j<k 0 ≤ i , j < k ,
U i j = U 2 ( i + j ) , P i j = V 2 ( i + j ) + 1 , Q i j = U 2 ( i + j ) + 1 , V i j = V 2 ( i + j ) + 2 . (52) \begin{aligned}
\mathcal{U}_{ij}&=U_{2(i+j)},&\mathcal{P}_{ij}&=V_{2(i+j)+1},\\
\mathcal{Q}_{ij}&=U_{2(i+j)+1},&\mathcal{V}_{ij}&=V_{2(i+j)+2}. \tag*{(52)}
\end{aligned} U ij Q ij = U 2 ( i + j ) , = U 2 ( i + j ) + 1 , P ij V ij = V 2 ( i + j ) + 1 , = V 2 ( i + j ) + 2 . (52)
Thus, the four sequences in the extended Cusick identity are:
u j = U 2 j , p j = V 2 j + 1 , q j = U 2 j + 1 , v j = V 2 j + 2 . (53) u_j=U_{2j},\qquad p_j=V_{2j+1},\qquad q_j=U_{2j+1},\qquad v_j=V_{2j+2}. \tag*{(53)} u j = U 2 j , p j = V 2 j + 1 , q j = U 2 j + 1 , v j = V 2 j + 2 . (53)
Then, applying the extended Cusick identity and using (43) , we find
det T 2 k ( Φ ′ ) = pf [ C ( u , v ) − C ( q , p ) ] = pf [ 2 Λ ( m ) ω ( s ) ( Λ ( m ) ) T ] = 2 k det ( Λ ( m ) ) pf ( ω ( s ) ) . (54) \det\mathcal{T}_{2k}^{(\Phi')}=\operatorname{pf}\left[C(u,v)-C(q,p)\right]=\operatorname{pf}\left[2\Lambda^{(m)}\omega^{(s)}\left(\Lambda^{(m)}\right)^{\mathsf{T}}\right]=2^k\det\left(\Lambda^{(m)}\right)\operatorname{pf}\left(\omega^{(s)}\right). \tag*{(54)} det T 2 k ( Φ ′ ) = pf [ C ( u , v ) − C ( q , p ) ] = pf [ 2 Λ ( m ) ω ( s ) ( Λ ( m ) ) T ] = 2 k det ( Λ ( m ) ) pf ( ω ( s ) ) . (54)
Odd case (m = 2 k + 1 m=2k+1 m = 2 k + 1 ). The corresponding Toda matrix has the form:
T 2 k + 1 ( Φ ′ ) = ( U 0 V 1 U 2 ⋯ U m − 1 U 1 V 2 U 3 ⋯ U m ⋮ ⋮ ⋮ ⋱ ⋮ U m − 1 V m U m + 1 ⋯ U 2 m − 2 ) , m = 2 k + 1. (55) \mathcal{T}_{2k+1}^{(\Phi')}=
\begin{pmatrix}
U_0&V_1&U_2&\cdots&U_{m-1}\\
U_1&V_2&U_3&\cdots&U_m\\
\vdots&\vdots&\vdots&\ddots&\vdots\\
U_{m-1}&V_m&U_{m+1}&\cdots&U_{2m-2}
\end{pmatrix},
\qquad m=2k+1. \tag*{(55)} T 2 k + 1 ( Φ ′ ) = U 0 U 1 ⋮ U m − 1 V 1 V 2 ⋮ V m U 2 U 3 ⋮ U m + 1 ⋯ ⋯ ⋱ ⋯ U m − 1 U m ⋮ U 2 m − 2 , m = 2 k + 1. (55)
The definitions of u u u , p p p , q q q , v v v and the relation (43) are valid for arbitrary m m m , so we now use them with m = 2 k + 1 m=2k+1 m = 2 k + 1 . Then as in the even case, by elementary row and column permutations, the Toda matrix T m ( Φ ′ ) \mathcal{T}_{m}^{(\Phi')} T m ( Φ ′ ) can be brought into the block-Hankel form. Applying the bordered extended Cusick identity, we obtain
det T m ( Φ ′ ) = det ( U P Q V ) = pf ( C ( u , v ) − C ( q , p ) u − u T 0 ) (56) \det\mathcal{T}_{m}^{(\Phi')}=\det\begin{pmatrix}\mathcal{U}&\mathcal{P}\\\mathcal{Q}&\mathcal{V}\end{pmatrix}=\operatorname{pf}\begin{pmatrix}C(u,v)-C(q,p)&u\\-\mathbf{u}^{\mathsf{T}}&0\end{pmatrix} \tag*{(56)} det T m ( Φ ′ ) = det ( U Q P V ) = pf ( C ( u , v ) − C ( q , p ) − u T u 0 ) (56)
where u = Λ ( m ) Γ ( + ) = ( u 0 , … , u m − 1 ) T \mathbf{u}=\Lambda^{(m)}\boldsymbol{\Gamma}^{(+)}=(u_0,\ldots,u_{m-1})^{\mathsf{T}} u = Λ ( m ) Γ ( + ) = ( u 0 , … , u m − 1 ) T . Using (43) , we find:
det T m ( Φ ′ ) = pf ( 2 Λ ( m ) ω ( s ) ( Λ ( m ) ) T Λ ( m ) Γ ( + ) − ( Γ ( + ) ) T ( Λ ( m ) ) T 0 ) . (57) \det\mathcal{T}_{m}^{(\Phi')}=\operatorname{pf}\begin{pmatrix}2\Lambda^{(m)}\omega^{(s)}(\Lambda^{(m)})^{\mathsf{T}}&\Lambda^{(m)}\boldsymbol{\Gamma}^{(+)}\\-(\boldsymbol{\Gamma}^{(+)})^{\mathsf{T}}(\Lambda^{(m)})^{\mathsf{T}}&0\end{pmatrix}. \tag*{(57)} det T m ( Φ ′ ) = pf ( 2 Λ ( m ) ω ( s ) ( Λ ( m ) ) T − ( Γ ( + ) ) T ( Λ ( m ) ) T Λ ( m ) Γ ( + ) 0 ) . (57)
Defining D = diag ( Λ ( m ) , 1 ) D=\operatorname{diag}(\Lambda^{(m)},1) D = diag ( Λ ( m ) , 1 ) , we obtain:
det T m ( Φ ′ ) = pf [ D ( 2 ω ( s ) Γ ( + ) − ( Γ ( + ) ) T 0 ) D T ] = 2 k det Λ ( m ) pf ( ω ( s ) Γ ( + ) − ( Γ ( + ) ) T 0 ) . (58) \det\mathcal{T}_{m}^{(\Phi')}=\operatorname{pf}\left[D\begin{pmatrix}2\omega^{(s)}&\boldsymbol{\Gamma}^{(+)}\\-(\boldsymbol{\Gamma}^{(+)})^{\mathsf{T}}&0\end{pmatrix}D^{\mathsf{T}}\right]=2^{k}\det\Lambda^{(m)}\operatorname{pf}\begin{pmatrix}\omega^{(s)}&\boldsymbol{\Gamma}^{(+)}\\-(\boldsymbol{\Gamma}^{(+)})^{\mathsf{T}}&0\end{pmatrix}. \tag*{(58)} det T m ( Φ ′ ) = pf [ D ( 2 ω ( s ) − ( Γ ( + ) ) T Γ ( + ) 0 ) D T ] = 2 k det Λ ( m ) pf ( ω ( s ) − ( Γ ( + ) ) T Γ ( + ) 0 ) . (58)
Using the relation between ω \omega ω and ω ( s ) \omega^{(s)} ω ( s ) :
ω = ω ( s ) + 1 2 [ Γ ( x ) ( Γ ( + ) ) T − Γ ( + ) ( Γ ( x ) ) T ] , (59) \omega=\omega^{(s)}+\frac{1}{2}\left[\boldsymbol{\Gamma}^{(x)}(\boldsymbol{\Gamma}^{(+)})^{\mathsf{T}}-\boldsymbol{\Gamma}^{(+)}(\boldsymbol{\Gamma}^{(x)})^{\mathsf{T}}\right], \tag*{(59)} ω = ω ( s ) + 2 1 [ Γ ( x ) ( Γ ( + ) ) T − Γ ( + ) ( Γ ( x ) ) T ] , (59)
and introducing the triangular matrix:
L = ( I m − 1 2 Γ ( x ) 0 0 1 ) , det L = 1 , (60) L=\begin{pmatrix}I_m-\frac{1}{2}\boldsymbol{\Gamma}^{(x)}&0\\0&1\end{pmatrix},\qquad\det L=1, \tag*{(60)} L = ( I m − 2 1 Γ ( x ) 0 0 1 ) , det L = 1 , (60)
we obtain:
ω ^ ( s ) = L ( ω ( s ) Γ ( + ) − ( Γ ( + ) ) T 0 ) L T . (61) \widehat{\omega}^{(s)}=L\begin{pmatrix}\omega^{(s)}&\boldsymbol{\Gamma}^{(+)}\\-(\boldsymbol{\Gamma}^{(+)})^{\mathsf{T}}&0\end{pmatrix}L^{\mathsf{T}}. \tag*{(61)} ω ( s ) = L ( ω ( s ) − ( Γ ( + ) ) T Γ ( + ) 0 ) L T . (61)
Since det L = 1 \det L=1 det L = 1 , the Pfaffian identity pf ( M A M T ) = det ( M ) pf ( A ) \operatorname{pf}(MAM^{\mathsf{T}})=\det(M)\operatorname{pf}(A) pf ( M A M T ) = det ( M ) pf ( A ) gives:
pf ( ω ^ ( s ) ) = pf ( ω ( s ) Γ ( + ) − ( Γ ( + ) ) T 0 ) . (62) \operatorname{pf}(\widehat{\omega}^{(s)})=\operatorname{pf}\begin{pmatrix}\omega^{(s)}&\boldsymbol{\Gamma}^{(+)}\\-(\boldsymbol{\Gamma}^{(+)})^{\mathsf{T}}&0\end{pmatrix}. \tag*{(62)} pf ( ω ( s ) ) = pf ( ω ( s ) − ( Γ ( + ) ) T Γ ( + ) 0 ) . (62)
Therefore,
det T m ( Φ ′ ) = 2 k det Λ ( m ) pf ( ω ^ ( s ) ) , m = 2 k + 1. (63) \det\mathcal{T}_{m}^{(\Phi')}=2^{k}\det\Lambda^{(m)}\operatorname{pf}(\widehat{\omega}^{(s)}),\qquad m=2k+1. \tag*{(63)} det T m ( Φ ′ ) = 2 k det Λ ( m ) pf ( ω ( s ) ) , m = 2 k + 1. (63)
Next, we prove analogous results for the seed Φ \Phi Φ .
Proof for the seed Φ \Phi Φ
Lemma 7. Let m = n − 1 m=n-1 m = n − 1 , and consider the Toda τ \tau τ -function associated with the seed Φ \Phi Φ . Then
det T m + 1 ( Φ ) = { 2 k ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( e ) ) , m = 2 k , 2 k + 1 ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( o ) ) , m = 2 k + 1 , (64) \det\mathcal{T}_{m+1}^{(\Phi)}=
\begin{cases}
2^{k}(2m+1)\det\Lambda^{(m)}\operatorname{pf}\left(\Omega^{(e)}\right),&m=2k,\\
2^{k+1}(2m+1)\det\Lambda^{(m)}\operatorname{pf}\left(\Omega^{(o)}\right),&m=2k+1,
\end{cases} \tag*{(64)} det T m + 1 ( Φ ) = { 2 k ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( e ) ) , 2 k + 1 ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( o ) ) , m = 2 k , m = 2 k + 1 , (64)
where Λ ( m ) \Lambda^{(m)} Λ ( m ) is defined in (30) , and Ω ( e ) , Ω ( o ) \Omega^{(e)},\Omega^{(o)} Ω ( e ) , Ω ( o ) are given in Appendix A.
Proof. Even case (m = 2 k m=2k m = 2 k ). For the seed Φ = − p x + i q y \Phi=-px+iqy Φ = − p x + i q y , we introduce the sequences
A j = P j ( x ) − i s P j ( y ) , B j = P j ( x ) + i s P j ( y ) . (65) A_j=P_j(x)-isP_j(y),\qquad B_j=P_j(x)+isP_j(y). \tag*{(65)} A j = P j ( x ) − i s P j ( y ) , B j = P j ( x ) + i s P j ( y ) . (65)
The corresponding normalized Toda matrix has the form:
T m + 1 ( Φ ) = ( A 0 B 1 A 2 ⋯ A m A 1 B 2 A 3 ⋯ A m + 1 ⋮ ⋮ ⋮ ⋱ ⋮ A m B m + 1 A m + 2 ⋯ A 2 m ) , m = 2 k . (66) \mathcal{T}_{m+1}^{(\Phi)}=
\begin{pmatrix}
A_0 & B_1 & A_2 & \cdots& A_m\\
A_1 & B_2 & A_3 & \cdots& A_{m+1}\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
A_m & B_{m+1} & A_{m+2} & \cdots& A_{2m}
\end{pmatrix},
\qquad m=2k. \tag*{(66)} T m + 1 ( Φ ) = A 0 A 1 ⋮ A m B 1 B 2 ⋮ B m + 1 A 2 A 3 ⋮ A m + 2 ⋯ ⋯ ⋱ ⋯ A m A m + 1 ⋮ A 2 m , m = 2 k . (66)
As in the denominator case, by elementary row and column permutations, the Toda matrix can be brought into block-Hankel form:
det T m + 1 ( Φ ) = det ( U P Q V ) , (67) \det\mathcal{T}_{m+1}^{(\Phi)}
=\det\begin{pmatrix}
U & P\\
Q & V
\end{pmatrix}, \tag*{(67)} det T m + 1 ( Φ ) = det ( U Q P V ) , (67)
where
U i j = A 2 ( i + j ) , P i j = B 2 ( i + j ) + 1 , U_{ij}=A_{2(i+j)},\qquad P_{ij}=B_{2(i+j)+1}, U ij = A 2 ( i + j ) , P ij = B 2 ( i + j ) + 1 ,
Q i j = A 2 ( i + j ) + 1 , V i j = B 2 ( i + j ) + 2 , (68) Q_{ij}=A_{2(i+j)+1},\qquad V_{ij}=B_{2(i+j)+2}, \tag*{(68)} Q ij = A 2 ( i + j ) + 1 , V ij = B 2 ( i + j ) + 2 , (68)
with block dimensions ( k + 1 ) × ( k + 1 ) (k+1)\times(k+1) ( k + 1 ) × ( k + 1 ) , ( k + 1 ) × k (k+1)\times k ( k + 1 ) × k , k × ( k + 1 ) k\times(k+1) k × ( k + 1 ) , and k × k k\times k k × k , respectively. Thus, the four sequences entering the extended Cusick identity are
u j = A 2 j , p j = B 2 j + 1 , q j = A 2 j + 1 , v j = B 2 j + 2 . (69) u_j=A_{2j},\qquad p_j=B_{2j+1},\qquad q_j=A_{2j+1},\qquad v_j=B_{2j+2}. \tag*{(69)} u j = A 2 j , p j = B 2 j + 1 , q j = A 2 j + 1 , v j = B 2 j + 2 . (69)
Applying the bordered extended Cusick identity, we obtain
det T m + 1 ( Φ ) = pf ( C ( u , v ) − C ( q , p ) u − u T 0 ) , (70) \det\mathcal{T}_{m+1}^{(\Phi)}
=\operatorname{pf}\begin{pmatrix}
C(u,v)-C(q,p) & \mathbf{u}\\
-\mathbf{u}^{T} & 0
\end{pmatrix}, \tag*{(70)} det T m + 1 ( Φ ) = pf ( C ( u , v ) − C ( q , p ) − u T u 0 ) , (70)
where u = ( A 0 , A 2 , … , A 2 m ) T \mathbf{u}=(A_0,A_2,\ldots,A_{2m})^T u = ( A 0 , A 2 , … , A 2 m ) T .
To express this Pfaffian in terms of the Ernst matrices, we first note that the same calculation leading to (36) gives:8
[ C ( u , v ) − C ( q , p ) ] a b = ∑ ℓ = 0 2 ( b − a ) − 1 ( − 1 ) ℓ A 2 a − 2 + ℓ B 2 b − 2 − ℓ = I 2 a − 2 , 2 b − 2 ( P ) ( x ) + s 2 I 2 a − 2 , 2 b − 2 ( P ) ( y ) + i s [ P 2 a − 2 ( x ) P 2 b − 2 ( y ) − P 2 a − 2 ( y ) P 2 b − 2 ( x ) ] , 1 ≤ a < b ≤ m + 1. (71) \begin{aligned}
\left[C(u,v)-C(q,p)\right]_{ab}
=\sum_{\ell=0}^{2(b-a)-1}(-1)^\ell A_{2a-2+\ell}B_{2b-2-\ell} \\
&=\mathcal{I}^{(P)}_{2a-2,2b-2}(x)+s^2\mathcal{I}^{(P)}_{2a-2,2b-2}(y)
+is\left[P_{2a-2}(x)P_{2b-2}(y)-P_{2a-2}(y)P_{2b-2}(x)\right],
\qquad1\leq a<b\leq m+1. \tag*{(71)}
\end{aligned} [ C ( u , v ) − C ( q , p ) ] ab = ℓ = 0 ∑ 2 ( b − a ) − 1 ( − 1 ) ℓ A 2 a − 2 + ℓ B 2 b − 2 − ℓ = I 2 a − 2 , 2 b − 2 ( P ) ( x ) + s 2 I 2 a − 2 , 2 b − 2 ( P ) ( y ) + i s [ P 2 a − 2 ( x ) P 2 b − 2 ( y ) − P 2 a − 2 ( y ) P 2 b − 2 ( x ) ] , 1 ≤ a < b ≤ m + 1. (71)
∑ ℓ = 0 d − 1 ( − 1 ) ℓ P r + ℓ ( y ) P t − ℓ ( x ) = ∑ ℓ = 1 d ( − 1 ) ℓ P r + ℓ ( x ) P t − ℓ ( y ) . \sum_{\ell=0}^{d-1}(-1)^\ell P_{r+\ell}(y)P_{t-\ell}(x)
=\sum_{\ell=1}^{d}(-1)^\ell P_{r+\ell}(x)P_{t-\ell}(y). ℓ = 0 ∑ d − 1 ( − 1 ) ℓ P r + ℓ ( y ) P t − ℓ ( x ) = ℓ = 1 ∑ d ( − 1 ) ℓ P r + ℓ ( x ) P t − ℓ ( y ) .
Thus, since d d d is even, all intermediate terms cancel, leaving P r ( x ) P t ( y ) − P r ( y ) P t ( x ) P_r(x)P_t(y)-P_r(y)P_t(x) P r ( x ) P t ( y ) − P r ( y ) P t ( x ) . We have also used the fact that the identity (35) remains valid for r = 0 r=0 r = 0 , since P j ( 1 ) = 0 P_j(1)=0 P j ( 1 ) = 0 for j ≥ 1 j\geq1 j ≥ 1 , and therefore the boundary terms at z = 1 z=1 z = 1 vanish.
Note that unlike the denominator case, the polynomial sequence now begins with P 0 ( z ) = z P_0(z)=z P 0 ( z ) = z . From the definition of Γ j \Gamma_j Γ j , we have:
P 0 ( z ) = ∑ j = 1 m ( 2 j − 1 ) Γ j ( z ) + ( 2 m + 1 ) F ( z ) , F ( z ) = z 2 m + 1 2 m + 1 . (72) P_0(z)=\sum_{j=1}^{m}(2j-1)\Gamma_j(z)+(2m+1)F(z),\qquad F(z)=\frac{z^{2m+1}}{2m+1}. \tag*{(72)} P 0 ( z ) = j = 1 ∑ m ( 2 j − 1 ) Γ j ( z ) + ( 2 m + 1 ) F ( z ) , F ( z ) = 2 m + 1 z 2 m + 1 . (72)
Introducing the vector
d = ( 1 , 3 , … , 2 m − 1 ) T , (73) \mathbf{d}=(1,3,\ldots,2m-1)^T, \tag*{(73)} d = ( 1 , 3 , … , 2 m − 1 ) T , (73)
and using (29) , we obtain
\begin{pmatrix}
P_0\\
P_2\\
\vdots\\
P_{2m}
\end{pmatrix}
=M^{(m)}
\begin{pmatrix}
\Gamma_1\\
\vdots\\
\Gamma_m\\
F
\end{pmatrix},
\qquad
M^{(m)}=
\begin{pmatrix}
\mathbf{d}^T & 2m+1\\
\Lambda^{(m)} & \mathbf{0}
\end{pmatrix}. \tag*{(74)}\end{pmatrix}
Expanding along the last column, we find
det M ( m ) = ( − 1 ) m ( 2 m + 1 ) det Λ ( m ) . (75) \det M^{(m)}=(-1)^m(2m+1)\det\Lambda^{(m)}. \tag*{(75)} det M ( m ) = ( − 1 ) m ( 2 m + 1 ) det Λ ( m ) . (75)
Substituting the polynomial transformation (74) into (71) , the terms involving Γ i \Gamma_i Γ i and Γ j \Gamma_j Γ j give:
∫ 1 x ( Γ i Γ j ′ − Γ i ′ Γ j ) d w + s 2 ∫ 1 y ( Γ i Γ j ′ − Γ i ′ Γ j ) d w + i s [ Γ i ( x ) Γ j ( y ) − Γ i ( y ) Γ j ( x ) ] = 2 ω i j ( − s ) . (76) \int_1^x(\Gamma_i\Gamma_j'-\Gamma_i'\Gamma_j)\,dw+s^2\int_1^y(\Gamma_i\Gamma_j'-\Gamma_i'\Gamma_j)\,dw+is[\Gamma_i(x)\Gamma_j(y)-\Gamma_i(y)\Gamma_j(x)]
=2\omega_{ij}^{(-s)}. \tag*{(76)} ∫ 1 x ( Γ i Γ j ′ − Γ i ′ Γ j ) d w + s 2 ∫ 1 y ( Γ i Γ j ′ − Γ i ′ Γ j ) d w + i s [ Γ i ( x ) Γ j ( y ) − Γ i ( y ) Γ j ( x )] = 2 ω ij ( − s ) . (76)
Similarly, the terms involving Γ i \Gamma_i Γ i and F F F , after integration by parts using F ′ ( w ) = w 2 m F'(w)=w^{2m} F ′ ( w ) = w 2 m and Γ i ( 1 ) = 0 \Gamma_i(1)=0 Γ i ( 1 ) = 0 , give:
∫ 1 x ( Γ i F ′ − Γ i ′ F ) d w + s 2 ∫ 1 y ( Γ i F ′ − Γ i ′ F ) d w + i s [ Γ i ( x ) F ( y ) − Γ i ( y ) F ( x ) ] = 2 [ ∫ 1 x Γ i ( w ) w 2 m d w + s 2 ∫ 1 y Γ i ( w ) w 2 m d w ] − γ ‾ β i = 2 ρ i − γ ‾ β i = − 2 σ i , (77) \begin{aligned}
&\int_1^x(\Gamma_iF'-\Gamma_i'F)\,dw+s^2\int_1^y(\Gamma_iF'-\Gamma_i'F)\,dw+is[\Gamma_i(x)F(y)-\Gamma_i(y)F(x)]\\
&=2\left[\int_1^x\Gamma_i(w)w^{2m}\,dw+s^2\int_1^y\Gamma_i(w)w^{2m}\,dw\right]-\overline{\gamma}\beta_i\\
&=2\rho_i-\overline{\gamma}\beta_i=-2\sigma_i, \tag*{(77)}
\end{aligned} ∫ 1 x ( Γ i F ′ − Γ i ′ F ) d w + s 2 ∫ 1 y ( Γ i F ′ − Γ i ′ F ) d w + i s [ Γ i ( x ) F ( y ) − Γ i ( y ) F ( x )] = 2 [ ∫ 1 x Γ i ( w ) w 2 m d w + s 2 ∫ 1 y Γ i ( w ) w 2 m d w ] − γ β i = 2 ρ i − γ β i = − 2 σ i , (77)
where the quantities ρ i \rho_i ρ i , β i \beta_i β i , γ \gamma γ , and σ i \sigma_i σ i are defined in Appendix A. Thus, (71) takes the matrix form:
C ( u , v ) − C ( q , p ) = 2 M ( m ) ( ω ( − s ) − σ σ T 0 ) ( M ( m ) ) T , (78) C(u,v)-C(q,p)=2M^{(m)}
\begin{pmatrix}
\omega^{(-s)}&-\boldsymbol{\sigma}\\
\boldsymbol{\sigma}^T&0
\end{pmatrix}
(M^{(m)})^T, \tag*{(78)} C ( u , v ) − C ( q , p ) = 2 M ( m ) ( ω ( − s ) σ T − σ 0 ) ( M ( m ) ) T , (78)
where σ = ( σ 1 , … , σ m ) T \boldsymbol{\sigma}=(\sigma_1,\ldots,\sigma_m)^T σ = ( σ 1 , … , σ m ) T . The polynomial transformation (74) also gives the border vector:
u = M ( m ) ( β γ ) , β = ( β 1 , … , β m ) T . (79) \mathbf{u}=M^{(m)}
\begin{pmatrix}
\boldsymbol{\beta}\\
\gamma
\end{pmatrix},
\qquad
\boldsymbol{\beta}=(\beta_1,\ldots,\beta_m)^T. \tag*{(79)} u = M ( m ) ( β γ ) , β = ( β 1 , … , β m ) T . (79)
Introducing D = diag ( M ( m ) , 1 ) D=\operatorname{diag}(M^{(m)},1) D = diag ( M ( m ) , 1 ) , we then find:
det T m + 1 ( Φ ) = pf [ D ( 2 ω ( − s ) − 2 σ β 2 σ T 0 γ − β T − γ 0 ) D T ] = 2 k det M ( m ) pf ( ω ( − s ) − σ β σ T 0 γ − β T − γ 0 ) . (80) \det\mathcal{T}_{m+1}^{(\Phi)}
=\operatorname{pf}\left[
D
\begin{pmatrix}
2\omega^{(-s)}&-2\boldsymbol{\sigma}&\boldsymbol{\beta}\\
2\boldsymbol{\sigma}^T&0&\gamma\\
-\boldsymbol{\beta}^T&-\gamma&0
\end{pmatrix}
D^T
\right]
=2^k\det M^{(m)}\operatorname{pf}
\begin{pmatrix}
\omega^{(-s)}&-\boldsymbol{\sigma}&\boldsymbol{\beta}\\
\boldsymbol{\sigma}^T&0&\gamma\\
-\boldsymbol{\beta}^T&-\gamma&0
\end{pmatrix}. \tag*{(80)} det T m + 1 ( Φ ) = pf D 2 ω ( − s ) 2 σ T − β T − 2 σ 0 − γ β γ 0 D T = 2 k det M ( m ) pf ω ( − s ) σ T − β T − σ 0 − γ β γ 0 . (80)
Now, from the definitions of ω ( s ) \omega^{(s)} ω ( s ) and σ i \sigma_i σ i , we have
ω ( s ) = ω ( − s ) + 1 2 [ β ˉ β T − β β ˉ T ] , − σ = ρ − γ 2 β ˉ . (81) \omega^{(s)}=\omega^{(-s)}+\frac{1}{2}\left[\bar{\beta}\beta^{T}-\beta\bar{\beta}^{T}\right],\qquad-\sigma=\rho-\frac{\gamma}{2}\bar{\beta}. \tag*{(81)} ω ( s ) = ω ( − s ) + 2 1 [ β ˉ β T − β β ˉ T ] , − σ = ρ − 2 γ β ˉ . (81)
Introducing the triangular matrix
L = ( I m 0 − 1 2 β ˉ 0 1 0 0 0 1 ) , det L = 1 , (82) L=\begin{pmatrix} I_m & 0 & -\frac{1}{2}\bar{\beta} \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{pmatrix},\qquad\det L=1, \tag*{(82)} L = I m 0 0 0 1 0 − 2 1 β ˉ 0 1 , det L = 1 , (82)
we obtain
Ω ( e ) = L ( ω ( − s ) − σ β σ T 0 γ − β T − γ 0 ) L T . (83) \Omega^{(e)}=L\begin{pmatrix} \omega^{(-s)} & -\sigma& \beta\\ \sigma^{T} & 0 & \gamma\\ -\beta^{T} & -\gamma& 0 \end{pmatrix}L^{T}. \tag*{(83)} Ω ( e ) = L ω ( − s ) σ T − β T − σ 0 − γ β γ 0 L T . (83)
Since det L = 1 \det L=1 det L = 1 , the Pfaffian is unchanged by this transformation. Using (75) , we conclude that
det T m + 1 ( Φ ) = 2 k ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( e ) ) , m = 2 k . (84) \det T_{m+1}^{(\Phi)}=2^{k}(2m+1)\det\Lambda^{(m)}\operatorname{pf}\left(\Omega^{(e)}\right),\qquad m=2k. \tag*{(84)} det T m + 1 ( Φ ) = 2 k ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( e ) ) , m = 2 k . (84)
Odd case (m = 2 k + 1 m=2k+1 m = 2 k + 1 ). The Toda matrix now has even order m + 1 = 2 k + 2 m+1=2k+2 m + 1 = 2 k + 2 . Using the same definitions of u u u , p p p , q q q , v v v as in the even case, we can bring it into block-Hankel form, with all four blocks having size ( k + 1 ) × ( k + 1 ) (k+1)\times(k+1) ( k + 1 ) × ( k + 1 ) . Applying the extended Cusick identity and using (78) , we obtain:
det T m + 1 ( Φ ) = pf [ C ( u , v ) − C ( q , p ) ] = pf [ 2 M ( m ) ( ω ( − s ) − σ σ T 0 ) ( M ( m ) ) T ] = 2 k + 1 det M ( m ) pf ( ω ( − s ) − σ σ T 0 ) . (85) \begin{aligned}
\det T_{m+1}^{(\Phi)}&=\operatorname{pf}[C(u,v)-C(q,p)]\\
&=\operatorname{pf}\left[2M^{(m)}\begin{pmatrix}\omega^{(-s)}&-\sigma\\ \sigma^{T}&0\end{pmatrix}(M^{(m)})^{T}\right]\\
&=2^{k+1}\det M^{(m)}\operatorname{pf}\begin{pmatrix}\omega^{(-s)}&-\sigma\\ \sigma^{T}&0\end{pmatrix}. \tag*{(85)}
\end{aligned} det T m + 1 ( Φ ) = pf [ C ( u , v ) − C ( q , p )] = pf [ 2 M ( m ) ( ω ( − s ) σ T − σ 0 ) ( M ( m ) ) T ] = 2 k + 1 det M ( m ) pf ( ω ( − s ) σ T − σ 0 ) . (85)
The last matrix differs from Ω ( o ) \Omega^{(o)} Ω ( o ) only by the sign of its final row and column, so its Pfaffian is − pf ( Ω ( o ) ) -\operatorname{pf}\left(\Omega^{(o)}\right) − pf ( Ω ( o ) ) . Since m m m is odd, (75) also gives det M ( m ) = − ( 2 m + 1 ) det Λ ( m ) \det M^{(m)}=-(2m+1)\det\Lambda^{(m)} det M ( m ) = − ( 2 m + 1 ) det Λ ( m ) . Hence, we obtain:
det T m + 1 ( Φ ) = 2 k + 1 ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( o ) ) , m = 2 k + 1. (86) \det T_{m+1}^{(\Phi)}=2^{k+1}(2m+1)\det\Lambda^{(m)}\operatorname{pf}\left(\Omega^{(o)}\right),\qquad m=2k+1. \tag*{(86)} det T m + 1 ( Φ ) = 2 k + 1 ( 2 m + 1 ) det Λ ( m ) pf ( Ω ( o ) ) , m = 2 k + 1. (86)
Proof of Theorem 2 . Using the relation τ ℓ ( Ψ ) = ( − p ) ℓ det T ℓ ( Ψ ) \tau_{\ell}(\Psi)=(-p)^{\ell}\det T_{\ell}^{(\Psi)} τ ℓ ( Ψ ) = ( − p ) ℓ det T ℓ ( Ψ ) , the two lemmas give:
τ n ( Φ ) τ n − 1 ( Φ ′ ) = { − p ( 2 n − 1 ) pf ( Ω ( e ) ) pf ( ω ( s ) ) , m = 2 k , − 2 p ( 2 n − 1 ) pf ( Ω ( o ) ) pf ( ω ^ ( s ) ) , m = 2 k + 1. (87) \frac{\tau_n(\Phi)}{\tau_{n-1}(\Phi')}=
\begin{cases}
-p(2n-1)\dfrac{\operatorname{pf}\left(\Omega^{(e)}\right)}{\operatorname{pf}\left(\omega^{(s)}\right)}, & m=2k,\\
-2p(2n-1)\dfrac{\operatorname{pf}\left(\Omega^{(o)}\right)}{\operatorname{pf}\left(\widehat{\omega}^{(s)}\right)}, & m=2k+1.
\end{cases} \tag*{(87)} τ n − 1 ( Φ ′ ) τ n ( Φ ) = ⎩ ⎨ ⎧ − p ( 2 n − 1 ) pf ( ω ( s ) ) pf ( Ω ( e ) ) , − 2 p ( 2 n − 1 ) pf ( ω ( s ) ) pf ( Ω ( o ) ) , m = 2 k , m = 2 k + 1. (87)
Comparing these expressions with Theorem 1 , we obtain
τ n ( Φ ) τ n − 1 ( Φ ′ ) = − g n f n = − ξ n , (88) \frac{\tau_n(\Phi)}{\tau_{n-1}(\Phi')}=-\frac{g_n}{f_n}=-\xi_n, \tag*{(88)} τ n − 1 ( Φ ′ ) τ n ( Φ ) = − f n g n = − ξ n , (88)
which proves Nakamura’s conjecture.
Conclusion
We have proved Nakamura’s conjecture for the Yamazaki-Hori rational solutions of the Ernst equation for arbitrary n n n . The proof establishes explicit identities between the Toda τ \tau τ -functions associated with the seeds Φ \Phi Φ and Φ ′ \Phi' Φ ′ and the Pfaffians representing the corresponding Ernst solutions. These identities hold for general values of p , q p,q p , q satisfying p 2 + q 2 = 1 p^2+q^2=1 p 2 + q 2 = 1 .
The main algebraic ingredient is an extension of the generalized Cusick identity to four arbitrary sequences. This identity expresses the Pfaffian of an antisymmetric matrix formed from convolution sums as the determinant of a block-Hankel matrix. We have also established a bordered version, needed for the odd-dimensional Toda matrices. Both identities hold independently of the Ernst equation and may be useful in establishing other determinant–Pfaffian correspondences.
An interesting question is whether the Pfaffian representations obtained in our previous work can be shown to satisfy the Ernst equation directly, without using their equivalence to the Yamazaki-Hori solutions. This would require deriving differential identities for the Pfaffians and using them to establish the bilinear equations associated with the Ernst equation.
Another question is whether other choices of the seed function Φ \Phi Φ in the 2D Toda molecule lead to solutions of the Ernst equation. It would be interesting to identify the conditions on Φ \Phi Φ under which such a correspondence exists.
In this appendix, we collect the explicit formulas for the antisymmetric matrices appearing in the Pfaffian representations of the Ernst polynomials f n f_n f n and g n g_n g n , obtained in our previous work [1 ] .9
We begin with the matrices ω \omega ω and S S S , defined by:
ω i j : = 1 2 ∫ 1 x [ Γ i ( w ) Γ j ′ ( w ) − Γ i ′ ( w ) Γ j ( w ) ] d w + s 2 2 ∫ 1 y [ Γ i ( w ) Γ j ′ ( w ) − Γ i ′ ( w ) Γ j ( w ) ] d w , (89) \omega_{ij}:=\frac{1}{2}\int_1^x\left[\Gamma_i(w)\Gamma'_j(w)-\Gamma'_i(w)\Gamma_j(w)\right]\,dw+\frac{s^2}{2}\int_1^y\left[\Gamma_i(w)\Gamma'_j(w)-\Gamma'_i(w)\Gamma_j(w)\right]\,dw, \tag*{(89)} ω ij := 2 1 ∫ 1 x [ Γ i ( w ) Γ j ′ ( w ) − Γ i ′ ( w ) Γ j ( w ) ] d w + 2 s 2 ∫ 1 y [ Γ i ( w ) Γ j ′ ( w ) − Γ i ′ ( w ) Γ j ( w ) ] d w , (89)
S i j : = 1 2 Γ i ( x ) Γ j ( x ) + s 2 2 Γ i ( y ) Γ j ( y ) . (90) S_{ij}:=\frac{1}{2}\Gamma_i(x)\Gamma_j(x)+\frac{s^2}{2}\Gamma_i(y)\Gamma_j(y). \tag*{(90)} S ij := 2 1 Γ i ( x ) Γ j ( x ) + 2 s 2 Γ i ( y ) Γ j ( y ) . (90)
Here the polynomials Γ i \Gamma_i Γ i are given by:
Γ i ( w ) : = 1 2 i − 1 [ w 2 i − 1 − w 2 i + 1 ] . (91) \Gamma_i(w):=\frac{1}{2i-1}\left[w^{2i-1}-w^{2i+1}\right]. \tag*{(91)} Γ i ( w ) := 2 i − 1 1 [ w 2 i − 1 − w 2 i + 1 ] . (91)
We further define the antisymmetric matrix ω ( s ) \omega^{(s)} ω ( s ) :
ω i j ( s ) : = ω i j − i s Γ ^ i j , Γ ^ i j : = 1 2 [ Γ i ( x ) Γ j ( y ) − Γ i ( y ) Γ j ( x ) ] . (92) \omega^{(s)}_{ij}:=\omega_{ij}-\mathrm{i}s\,\widehat{\Gamma}_{ij},\qquad\widehat{\Gamma}_{ij}:=\frac{1}{2}\left[\Gamma_i(x)\Gamma_j(y)-\Gamma_i(y)\Gamma_j(x)\right]. \tag*{(92)} ω ij ( s ) := ω ij − i s Γ ij , Γ ij := 2 1 [ Γ i ( x ) Γ j ( y ) − Γ i ( y ) Γ j ( x ) ] . (92)
For odd m = n − 1 m=n-1 m = n − 1 , we define the bordered antisymmetric matrix:
ω ^ ( s ) = ( ω 11 ⋯ ω 1 m Γ 1 ( + ) ω 21 ⋯ ω 2 m Γ 2 ( + ) ⋮ ⋮ ⋱ ⋮ ω m 1 ⋯ ω m m Γ m ( + ) − Γ 1 ( + ) ⋯ − Γ m ( + ) 0 ) . (93) \widehat{\omega}^{(s)}=
\begin{pmatrix}
\omega_{11} & \cdots& \omega_{1m} & \Gamma^{(+)}_1\\
\omega_{21} & \cdots& \omega_{2m} & \Gamma^{(+)}_2\\
\vdots& \vdots& \ddots& \vdots\\
\omega_{m1} & \cdots& \omega_{mm} & \Gamma^{(+)}_m\\
-\Gamma^{(+)}_1 & \cdots& -\Gamma^{(+)}_m & 0
\end{pmatrix}. \tag*{(93)} ω ( s ) = ω 11 ω 21 ⋮ ω m 1 − Γ 1 ( + ) ⋯ ⋯ ⋮ ⋯ ⋯ ω 1 m ω 2 m ⋱ ω mm − Γ m ( + ) Γ 1 ( + ) Γ 2 ( + ) ⋮ Γ m ( + ) 0 . (93)
where
Γ j ( + ) ( x , y ) : = Γ j ( x ) + i s Γ j ( y ) . (94) \Gamma_j^{(+)}(x,y) := \Gamma_j(x) + \mathrm{i}s\Gamma_j(y). \tag*{(94)} Γ j ( + ) ( x , y ) := Γ j ( x ) + i s Γ j ( y ) . (94)
For even m = n − 1 m=n-1 m = n − 1 , the corresponding ( m + 2 ) × ( m + 2 ) (m+2)\times(m+2) ( m + 2 ) × ( m + 2 ) antisymmetric matrix entering the Pfaffian representation of g n g_n g n is:
Ω ( e ) = ( 0 ω 12 ( s ) ⋯ ω 1 m ( s ) ρ 1 β 1 − ω 12 ( s ) 0 ⋯ ω 2 m ( s ) ρ 2 β 2 ⋮ ⋮ ⋱ ⋮ ⋮ ⋮ − ω 1 m ( s ) − ω 2 m ( s ) ⋯ 0 ρ m β m − ρ 1 − ρ 2 ⋯ − ρ m 0 γ − β 1 − β 2 ⋯ − β m − γ 0 ) . (95) \Omega^{(e)} =
\begin{pmatrix}
0 & \omega_{12}^{(s)} & \cdots& \omega_{1m}^{(s)} & \rho_1 & \beta_1 \\
-\omega_{12}^{(s)} & 0 & \cdots& \omega_{2m}^{(s)} & \rho_2 & \beta_2 \\
\vdots& \vdots& \ddots& \vdots& \vdots& \vdots\\
-\omega_{1m}^{(s)} & -\omega_{2m}^{(s)} & \cdots& 0 & \rho_m & \beta_m \\
-\rho_1 & -\rho_2 & \cdots& -\rho_m & 0 & \gamma\\
-\beta_1 & -\beta_2 & \cdots& -\beta_m & -\gamma& 0
\end{pmatrix}. \tag*{(95)} Ω ( e ) = 0 − ω 12 ( s ) ⋮ − ω 1 m ( s ) − ρ 1 − β 1 ω 12 ( s ) 0 ⋮ − ω 2 m ( s ) − ρ 2 − β 2 ⋯ ⋯ ⋱ ⋯ ⋯ ⋯ ω 1 m ( s ) ω 2 m ( s ) ⋮ 0 − ρ m − β m ρ 1 ρ 2 ⋮ ρ m 0 − γ β 1 β 2 ⋮ β m γ 0 . (95)
For odd m = n − 1 m=n-1 m = n − 1 , the corresponding antisymmetric matrix has the form:
Ω ( o ) = ( 0 ω 12 ( − s ) ⋯ ω 1 m ( − s ) σ 1 − ω 12 ( − s ) 0 ⋯ ω 2 m ( − s ) σ 2 ⋮ ⋮ ⋱ ⋮ ⋮ − ω 1 m ( − s ) − ω 2 m ( − s ) ⋯ 0 σ m − σ 1 − σ 2 ⋯ − σ m 0 ) . (96) \Omega^{(o)} =
\begin{pmatrix}
0 & \omega_{12}^{(-s)} & \cdots& \omega_{1m}^{(-s)} & \sigma_1 \\
-\omega_{12}^{(-s)} & 0 & \cdots& \omega_{2m}^{(-s)} & \sigma_2 \\
\vdots& \vdots& \ddots& \vdots& \vdots\\
-\omega_{1m}^{(-s)} & -\omega_{2m}^{(-s)} & \cdots& 0 & \sigma_m \\
-\sigma_1 & -\sigma_2 & \cdots& -\sigma_m & 0
\end{pmatrix}. \tag*{(96)} Ω ( o ) = 0 − ω 12 ( − s ) ⋮ − ω 1 m ( − s ) − σ 1 ω 12 ( − s ) 0 ⋮ − ω 2 m ( − s ) − σ 2 ⋯ ⋯ ⋱ ⋯ ⋯ ω 1 m ( − s ) ω 2 m ( − s ) ⋮ 0 − σ m σ 1 σ 2 ⋮ σ m 0 . (96)
Here the quantities β j \beta_j β j , ρ j \rho_j ρ j , γ \gamma γ , and σ j \sigma_j σ j are defined by
β j = Γ j ( x ) − i s Γ j ( y ) , γ = x 2 m + 1 − i s y 2 m + 1 2 m + 1 , (97) \beta_j = \Gamma_j(x) - \mathrm{i}s\Gamma_j(y), \qquad\gamma= \frac{x^{2m+1}-\mathrm{i}s y^{2m+1}}{2m+1}, \tag*{(97)} β j = Γ j ( x ) − i s Γ j ( y ) , γ = 2 m + 1 x 2 m + 1 − i s y 2 m + 1 , (97)
ρ j = ∫ 1 x Γ j ( w ) w 2 m d w + s 2 ∫ 1 y Γ j ( w ) w 2 m d w , (98) \rho_j = \int_1^x \Gamma_j(w)w^{2m}\,\mathrm{d}w + s^2\int_1^y \Gamma_j(w)w^{2m}\,\mathrm{d}w, \tag*{(98)} ρ j = ∫ 1 x Γ j ( w ) w 2 m d w + s 2 ∫ 1 y Γ j ( w ) w 2 m d w , (98)
σ j = γ 2 β j − ρ j , 1 ≤ j ≤ m . (99) \sigma_j = \frac{\gamma}{2}\beta_j-\rho_j,\qquad1\leq j\leq m. \tag*{(99)} σ j = 2 γ β j − ρ j , 1 ≤ j ≤ m . (99)
Proofs of the extended Cusick identities
Proof of Theorem 4
We prove Theorem 4 by first establishing a convenient two-vector identity. Let
f r = ( f r ( 1 ) f r ( 2 ) ) , g r = ( g r ( 1 ) g r ( 2 ) ) , r ≥ 1 , (100) \mathbf{f}_r =
\begin{pmatrix}
f_r^{(1)} \\
f_r^{(2)}
\end{pmatrix},
\qquad
\mathbf{g}_r =
\begin{pmatrix}
g_r^{(1)} \\
g_r^{(2)}
\end{pmatrix},
\qquad r\geq1, \tag*{(100)} f r = ( f r ( 1 ) f r ( 2 ) ) , g r = ( g r ( 1 ) g r ( 2 ) ) , r ≥ 1 , (100)
be arbitrary sequences of two-dimensional vectors, and let A = A ( f , g ) \mathbf{A}=\mathbf{A}(\mathbf{f},\mathbf{g}) A = A ( f , g ) be the 2 n × 2 n 2n\times2n 2 n × 2 n antisymmetric matrix with entries
[ A ( f , g ) ] i j = { ∑ s = 1 j − i f s + i − 1 ⋅ g j − s , i < j , 0 , i = j , − [ A ( f , g ) ] j i , i > j . (101) [\mathbf{A}(\mathbf{f},\mathbf{g})]_{ij} =
\begin{cases}
\sum_{s=1}^{j-i}\mathbf{f}_{s+i-1}\cdot\mathbf{g}_{j-s}, & i<j,\\
0, & i=j,\\
-[\mathbf{A}(\mathbf{f},\mathbf{g})]_{ji}, & i>j.
\end{cases} \tag*{(101)} [ A ( f , g ) ] ij = ⎩ ⎨ ⎧ ∑ s = 1 j − i f s + i − 1 ⋅ g j − s , 0 , − [ A ( f , g ) ] j i , i < j , i = j , i > j . (101)
where
f i ⋅ g j = f i ( 1 ) g j ( 1 ) + f i ( 2 ) g j ( 2 ) (102) \mathbf{f}_i \cdot\mathbf{g}_j = f_i^{(1)}g_j^{(1)} + f_i^{(2)}g_j^{(2)} \tag*{(102)} f i ⋅ g j = f i ( 1 ) g j ( 1 ) + f i ( 2 ) g j ( 2 ) (102)
is the standard bilinear pairing. Also define
H f ( a ) = ( f r + s − 1 ( a ) ) r , s = 1 n , H g ( a ) = ( g r + s − 1 ( a ) ) r , s = 1 n , a = 1 , 2. (103) H_{\mathbf{f}}^{(a)} = \left(f_{r+s-1}^{(a)}\right)_{r,s=1}^{n}, \qquad H_{\mathbf{g}}^{(a)} = \left(g_{r+s-1}^{(a)}\right)_{r,s=1}^{n}, \qquad a = 1,2. \tag*{(103)} H f ( a ) = ( f r + s − 1 ( a ) ) r , s = 1 n , H g ( a ) = ( g r + s − 1 ( a ) ) r , s = 1 n , a = 1 , 2. (103)
We claim that
Φ ( f , g ) : = pf A ( f , g ) − det ( H f ( 1 ) H g ( 2 ) − H f ( 2 ) H g ( 1 ) ) ≡ 0. (104) \Phi(\mathbf{f},\mathbf{g}) := \operatorname{pf} A(\mathbf{f},\mathbf{g}) - \det\begin{pmatrix} H_{\mathbf{f}}^{(1)} & H_{\mathbf{g}}^{(2)} \\ -H_{\mathbf{f}}^{(2)} & H_{\mathbf{g}}^{(1)} \end{pmatrix} \equiv0. \tag*{(104)} Φ ( f , g ) := pf A ( f , g ) − det ( H f ( 1 ) − H f ( 2 ) H g ( 2 ) H g ( 1 ) ) ≡ 0. (104)
We prove this below by polarization and atomic parametrization. For the algebraic arguments below, let K \mathbb{K} K be a field of characteristic zero. All sequences, vectors, and matrices in this appendix are understood to have entries in K \mathbb{K} K . The application to the Ernst problem is obtained by taking K = C \mathbb{K} = \mathbb{C} K = C .
Proof of the auxiliary vector identity. The proof is based on two ingredients.
First, the matrix A ( f , g ) A(\mathbf{f},\mathbf{g}) A ( f , g ) is bilinear in f \mathbf{f} f and g \mathbf{g} g . Consequently, pf A ( f , g ) \operatorname{pf} A(\mathbf{f},\mathbf{g}) pf A ( f , g ) is bihomogeneous of degree ( n , n ) (n,n) ( n , n ) in ( f , g ) (\mathbf{f},\mathbf{g}) ( f , g ) . The determinant in the second term of (B.5) has the same bidegree: its first n n n columns are linear in f \mathbf{f} f , while its last n n n columns are linear in g \mathbf{g} g . Thus Φ \Phi Φ is bihomogeneous of bidegree ( n , n ) (n,n) ( n , n ) .
We define its full polarization by
Φ ~ ( F ( 1 ) , … , F ( n ) ; G ( 1 ) , … , G ( n ) ) : = [ t 1 ⋯ t n s 1 ⋯ s n ] Φ ( ∑ α = 1 n t α F ( α ) , ∑ β = 1 n s β G ( β ) ) , (105) \widetilde{\Phi}\left(\mathbf{F}_{(1)},\ldots,\mathbf{F}_{(n)};\mathbf{G}_{(1)},\ldots,\mathbf{G}_{(n)}\right) := [t_1\cdots t_n s_1\cdots s_n]\Phi\left(\sum_{\alpha=1}^{n} t_\alpha\mathbf{F}_{(\alpha)}, \sum_{\beta=1}^{n} s_\beta\mathbf{G}_{(\beta)}\right), \tag*{(105)} Φ ( F ( 1 ) , … , F ( n ) ; G ( 1 ) , … , G ( n ) ) := [ t 1 ⋯ t n s 1 ⋯ s n ] Φ α = 1 ∑ n t α F ( α ) , β = 1 ∑ n s β G ( β ) , (105)
where [ t 1 ⋯ t n s 1 ⋯ s n ] [t_1\cdots t_n s_1\cdots s_n] [ t 1 ⋯ t n s 1 ⋯ s n ] denotes the coefficient of the product of all the distinct parameters. The polarized form is multilinear in all of its 2 n 2n 2 n arguments, and
Φ ~ ( f , … , f ; g , … , g ) = ( n ! ) 2 Φ ( f , g ) . (106) \widetilde{\Phi}(\mathbf{f},\ldots,\mathbf{f};\mathbf{g},\ldots,\mathbf{g}) = (n!)^2\Phi(\mathbf{f},\mathbf{g}). \tag*{(106)} Φ ( f , … , f ; g , … , g ) = ( n ! ) 2 Φ ( f , g ) . (106)
Consequently, it is enough to prove that Φ ~ ≡ 0 \widetilde{\Phi} \equiv0 Φ ≡ 0 .
The second ingredient is the moment atomic representation. Put m = 2 n − 1 m = 2n - 1 m = 2 n − 1 , and choose arbitrary distinct nodes z 1 , … , z m z_1,\ldots,z_m z 1 , … , z m . The Vandermonde matrix
V ( z ) = ( z r k − 1 ) k , r = 1 m (107) V(z) = \left(z_r^{k-1}\right)_{k,r=1}^{m} \tag*{(107)} V ( z ) = ( z r k − 1 ) k , r = 1 m (107)
is invertible. Therefore every vector-valued moment sequence { f k } k = 1 m \{\mathbf{f}_k\}_{k=1}^{m} { f k } k = 1 m has an atomic representation
f k = ∑ r = 1 m p r z r k − 1 , (108) \mathbf{f}_k = \sum_{r=1}^{m} \mathbf{p}_r z_r^{k-1}, \tag*{(108)} f k = r = 1 ∑ m p r z r k − 1 , (108)
with uniquely determined vector weights p r \mathbf{p}_r p r , for the chosen nodes. Hence the vector atoms
a ( x , u ) : = ( u , u x , … , u x 2 n − 2 ) , u ∈ K 2 , (109) \mathbf{a}(x,\mathbf{u}) := \left(\mathbf{u},\mathbf{u}x,\ldots,\mathbf{u}x^{2n-2}\right), \qquad\mathbf{u} \in\mathbb{K}^{2}, \tag*{(109)} a ( x , u ) := ( u , u x , … , u x 2 n − 2 ) , u ∈ K 2 , (109)
span the space of all moment sequences. Since Φ ~ \widetilde{\Phi} Φ is multilinear, it is therefore enough to evaluate it when each of its arguments is one atom.
Choose atoms a ( x α , u α ) \mathbf{a}(x_\alpha,\mathbf{u}_\alpha) a ( x α , u α ) , a ( y β , v β ) \mathbf{a}(y_\beta,\mathbf{v}_\beta) a ( y β , v β ) , α , β = 1 , … , n \alpha,\beta=1,\ldots,n α , β = 1 , … , n , and define
F k ( t ) : = ∑ α = 1 n t α u α x α k − 1 , G k ( s ) : = ∑ β = 1 n s β v β y β k − 1 . (110) F_k(\mathbf{t}) := \sum_{\alpha=1}^{n} t_\alpha\mathbf{u}_\alpha x_\alpha^{k-1}, \qquad G_k(\mathbf{s}) := \sum_{\beta=1}^{n} s_\beta\mathbf{v}_\beta y_\beta^{k-1}. \tag*{(110)} F k ( t ) := α = 1 ∑ n t α u α x α k − 1 , G k ( s ) := β = 1 ∑ n s β v β y β k − 1 . (110)
It is enough to prove that Φ ( F ( t ) , G ( s ) ) \Phi(\mathbf{F}(\mathbf{t}),\mathbf{G}(\mathbf{s})) Φ ( F ( t ) , G ( s )) vanishes identically in the parameters t α , s β t_\alpha,s_\beta t α , s β . The coefficient of t 1 ⋯ t n s 1 ⋯ s n t_1\cdots t_n s_1\cdots s_n t 1 ⋯ t n s 1 ⋯ s n will then give the desired polarized identity.
For i < j i<j i < j , we have
[ A ( F ( t ) , G ( s ) ) ] i j = ∑ α , β = 1 n t α s β ( u α ⋅ v β ) ∑ r = 1 j − i x α r + i − 2 y β j − r − 1 = ∑ α , β = 1 n t α s β ( u α ⋅ v β ) x α i − 1 y β j − 1 − x α j − 1 y β i − 1 y β − x α , (111) \begin{aligned}
[\mathbf{A}(\mathbf{F}(\mathbf{t}),\mathbf{G}(\mathbf{s}))]_{ij}
=
\sum_{\alpha,\beta=1}^{n} t_\alpha s_\beta(\mathbf{u}_\alpha\cdot\mathbf{v}_\beta)
\sum_{r=1}^{j-i}x_\alpha^{r+i-2}y_\beta^{j-r-1} \\
&=
\sum_{\alpha,\beta=1}^{n}t_\alpha s_\beta(\mathbf{u}_\alpha\cdot\mathbf{v}_\beta)
\frac{x_\alpha^{i-1}y_\beta^{j-1}-x_\alpha^{j-1}y_\beta^{i-1}}{y_\beta-x_\alpha}, \tag*{(111)}
\end{aligned} [ A ( F ( t ) , G ( s )) ] ij = α , β = 1 ∑ n t α s β ( u α ⋅ v β ) r = 1 ∑ j − i x α r + i − 2 y β j − r − 1 = α , β = 1 ∑ n t α s β ( u α ⋅ v β ) y β − x α x α i − 1 y β j − 1 − x α j − 1 y β i − 1 , (111)
where in the last equality we used the sum of a geometric progression.
Since the atom weights are arbitrary, we may absorb the parameters t α t_\alpha t α and s β s_\beta s β into them for the generic calculation below. After the resulting identity has been established for arbitrary weights, we restore the original parameters and extract the coefficient of t 1 ⋯ t n s 1 ⋯ s n t_1\cdots t_n s_1\cdots s_n t 1 ⋯ t n s 1 ⋯ s n . Thus, from now on, write
t α u α = ( u α ( 1 ) u α ( 2 ) ) , s β v β = ( v β ( 1 ) v β ( 2 ) ) . (112) t_\alpha\mathbf{u}_\alpha=
\begin{pmatrix}
u_\alpha^{(1)}\\
u_\alpha^{(2)}
\end{pmatrix},
\qquad
s_\beta\mathbf{v}_\beta=
\begin{pmatrix}
v_\beta^{(1)}\\
v_\beta^{(2)}
\end{pmatrix}. \tag*{(112)} t α u α = ( u α ( 1 ) u α ( 2 ) ) , s β v β = ( v β ( 1 ) v β ( 2 ) ) . (112)
and relabel these absorbed weights again as u α \mathbf{u}_\alpha u α and v β \mathbf{v}_\beta v β . Define the diagonal matrices
P a : = diag ( u 1 ( a ) , … , u n ( a ) ) , Q a : = diag ( v 1 ( a ) , … , v n ( a ) ) , a = 1 , 2 , (113) P_a:=\operatorname{diag}(u_1^{(a)},\ldots,u_n^{(a)}),
\qquad
Q_a:=\operatorname{diag}(v_1^{(a)},\ldots,v_n^{(a)}),
\qquad a=1,2, \tag*{(113)} P a := diag ( u 1 ( a ) , … , u n ( a ) ) , Q a := diag ( v 1 ( a ) , … , v n ( a ) ) , a = 1 , 2 , (113)
as well as the Cauchy matrix
C α β : = 1 y β − x α . (114) C_{\alpha\beta}:=\frac{1}{y_\beta-x_\alpha}. \tag*{(114)} C α β := y β − x α 1 . (114)
Then
Γ α β : = u α ⋅ v β y β − x α (115) \Gamma_{\alpha\beta}:=\frac{\mathbf{u}_\alpha\cdot\mathbf{v}_\beta}{y_\beta-x_\alpha} \tag*{(115)} Γ α β := y β − x α u α ⋅ v β (115)
can be written as
Γ = P 1 C Q 1 + P 2 C Q 2 . (116) \Gamma=P_1CQ_1+P_2CQ_2. \tag*{(116)} Γ = P 1 C Q 1 + P 2 C Q 2 . (116)
Introduce the n × n n\times n n × n Vandermonde matrices
( X ) r α : = x α r − 1 , ( Y ) r β : = y β r − 1 , r = 1 , … , n , (117) (X)_{r\alpha}:=x_\alpha^{r-1},
\qquad
(Y)_{r\beta}:=y_\beta^{r-1},
\qquad r=1,\ldots,n, \tag*{(117)} ( X ) r α := x α r − 1 , ( Y ) r β := y β r − 1 , r = 1 , … , n , (117)
and the 2 n × n 2n\times n 2 n × n extended Vandermonde matrices
( X ^ ) i α : = x α i − 1 , ( Y ^ ) i β : = y β i − 1 , i = 1 , … , 2 n . (118) (\widehat{X})_{i\alpha}:=x_\alpha^{i-1},
\qquad
(\widehat{Y})_{i\beta}:=y_\beta^{i-1},
\qquad i=1,\ldots,2n. \tag*{(118)} ( X ) i α := x α i − 1 , ( Y ) i β := y β i − 1 , i = 1 , … , 2 n . (118)
Finally, put
W : = ( X ^ Y ^ ) . (119) \mathcal{W}:=(\widehat{X}\ \widehat{Y}). \tag*{(119)} W := ( X Y ) . (119)
Using (B.12), we obtain
A ( F , G ) = W K W T , where K : = ( 0 Γ − Γ T 0 ) . (120) \mathbf{A}(\mathbf{F},\mathbf{G})=\mathcal{W}\mathcal{K}\mathcal{W}^{\mathsf{T}},
\qquad
\text{where}\qquad
\mathcal{K}:=
\begin{pmatrix}
0 & \Gamma\\
-\Gamma^{\mathsf{T}} & 0
\end{pmatrix}. \tag*{(120)} A ( F , G ) = W K W T , where K := ( 0 − Γ T Γ 0 ) . (120)
Therefore, using the standard Pfaffian congruence property,
pf A = det W pf K = ( − 1 ) n ( n − 1 ) 2 det W det Γ . (121) \operatorname{pf} A = \det\mathcal{W}\operatorname{pf} K = (-1)^{\frac{n(n-1)}{2}}\det\mathcal{W}\det\Gamma. \tag*{(121)} pf A = det W pf K = ( − 1 ) 2 n ( n − 1 ) det W det Γ. (121)
Let
Δ ( x ) : = ∏ α < α ′ ( x α ′ − x α ) , Δ ( y ) : = ∏ β < β ′ ( y β ′ − y β ) . (122) \Delta(x) := \prod_{\alpha<\alpha'}(x_{\alpha'}-x_\alpha), \qquad\Delta(y) := \prod_{\beta<\beta'}(y_{\beta'}-y_\beta). \tag*{(122)} Δ ( x ) := α < α ′ ∏ ( x α ′ − x α ) , Δ ( y ) := β < β ′ ∏ ( y β ′ − y β ) . (122)
Since all the x x x -columns precede all the y y y -columns, we get
det W = Δ ( x ) Δ ( y ) ∏ α , β = 1 n ( y β − x α ) . (123) \det\mathcal{W} = \Delta(x)\Delta(y)\prod_{\alpha,\beta=1}^{n}(y_\beta-x_\alpha). \tag*{(123)} det W = Δ ( x ) Δ ( y ) α , β = 1 ∏ n ( y β − x α ) . (123)
The Cauchy determinant formula, with our convention for C C C , is
det C = ( − 1 ) n ( n − 1 ) 2 Δ ( x ) Δ ( y ) ∏ α , β = 1 n ( y β − x α ) . (124) \det C = (-1)^{\frac{n(n-1)}{2}}\frac{\Delta(x)\Delta(y)}{\prod_{\alpha,\beta=1}^{n}(y_\beta-x_\alpha)}. \tag*{(124)} det C = ( − 1 ) 2 n ( n − 1 ) ∏ α , β = 1 n ( y β − x α ) Δ ( x ) Δ ( y ) . (124)
For the moment, assume that P 1 , Q 1 , X , Y P_1,Q_1,X,Y P 1 , Q 1 , X , Y , and C C C are invertible. Put
D : = P 1 − 1 P 2 , E : = Q 2 Q 1 − 1 . (125) D := P_1^{-1}P_2, \qquad E := Q_2Q_1^{-1}. \tag*{(125)} D := P 1 − 1 P 2 , E := Q 2 Q 1 − 1 . (125)
Then
det Γ = det ( P 1 C ( 1 + C − 1 D C E ) Q 1 ) = det P 1 det Q 1 det C det ( 1 + C − 1 D C E ) . (126) \det\Gamma= \det\left(P_1C(1+C^{-1}DCE)Q_1\right) = \det P_1\det Q_1\det C\det(1+C^{-1}DCE). \tag*{(126)} det Γ = det ( P 1 C ( 1 + C − 1 D C E ) Q 1 ) = det P 1 det Q 1 det C det ( 1 + C − 1 D C E ) . (126)
Combining the last three equations gives
pf A = Δ ( x ) 2 Δ ( y ) 2 det P 1 det Q 1 det ( 1 + C − 1 D C E ) . (127) \operatorname{pf} A = \Delta(x)^2\Delta(y)^2\det P_1\det Q_1\det(1+C^{-1}DCE). \tag*{(127)} pf A = Δ ( x ) 2 Δ ( y ) 2 det P 1 det Q 1 det ( 1 + C − 1 D C E ) . (127)
We now calculate the determinant in the second term of (B.5). By construction,
H f ( a ) = X P a X T , H g ( a ) = Y Q a Y T . (128) H_f^{(a)} = X P_a X^{T}, \qquad H_g^{(a)} = Y Q_a Y^{T}. \tag*{(128)} H f ( a ) = X P a X T , H g ( a ) = Y Q a Y T . (128)
Therefore
det ( H f ( 1 ) H g ( 2 ) − H f ( 2 ) H g ( 1 ) ) = det ( X ) 2 det ( Y ) 2 det R = Δ ( x ) 2 Δ ( y ) 2 det R , (129) \det\begin{pmatrix} H_f^{(1)} & H_g^{(2)} \\ -H_f^{(2)} & H_g^{(1)} \end{pmatrix}
= \det(X)^2\det(Y)^2\det R
= \Delta(x)^2\Delta(y)^2\det R, \tag*{(129)} det ( H f ( 1 ) − H f ( 2 ) H g ( 2 ) H g ( 1 ) ) = det ( X ) 2 det ( Y ) 2 det R = Δ ( x ) 2 Δ ( y ) 2 det R , (129)
where
R : = ( P 1 X − 1 Y Q 2 − Y − 1 X P 2 Q 1 ) . (130) R := \begin{pmatrix} P_1 & X^{-1}YQ_2 \\ -Y^{-1}XP_2 & Q_1 \end{pmatrix}. \tag*{(130)} R := ( P 1 − Y − 1 X P 2 X − 1 Y Q 2 Q 1 ) . (130)
It remains to compare det R \det R det R with the last factor in (B.28). Set
S : = Y − 1 X , Λ : = diag ( λ 1 , … , λ n ) , λ β : = ∏ α = 1 n ( y β − x α ) , (131) S := Y^{-1}X, \qquad\Lambda:= \operatorname{diag}(\lambda_1,\ldots,\lambda_n), \qquad\lambda_\beta:= \prod_{\alpha=1}^{n}(y_\beta-x_\alpha), \tag*{(131)} S := Y − 1 X , Λ := diag ( λ 1 , … , λ n ) , λ β := α = 1 ∏ n ( y β − x α ) , (131)
where S β γ S_{\beta\gamma} S β γ are the Lagrange interpolation weights ℓ β ( z ) \ell_\beta(z) ℓ β ( z ) , evaluated at z = x γ z=x_\gamma z = x γ
ℓ β ( z ) : = ∏ δ ≠ β z − y δ y β − y δ , S β γ = ∏ δ ≠ β x γ − y δ y β − y δ ≡ ℓ β ( x γ ) . (132) \ell_\beta(z) := \prod_{\delta\ne\beta}\frac{z-y_\delta}{y_\beta-y_\delta}, \qquad S_{\beta\gamma} = \prod_{\delta\ne\beta}\frac{x_\gamma-y_\delta}{y_\beta-y_\delta} \equiv\ell_\beta(x_\gamma). \tag*{(132)} ℓ β ( z ) := δ = β ∏ y β − y δ z − y δ , S β γ = δ = β ∏ y β − y δ x γ − y δ ≡ ℓ β ( x γ ) . (132)
By the Lagrange interpolation formula
p α ( z ) : = ∏ ρ ≠ α ( z − x ρ ) = ∑ β = 1 n p α ( y β ) ℓ β ( z ) ≡ ∑ β = 1 n ∏ ρ ≠ α ( y β − x ρ ) ℓ β ( z ) . (133) p_{\alpha}(z) := \prod_{\rho\ne\alpha}(z-x_{\rho}) = \sum_{\beta=1}^{n}p_{\alpha}(y_{\beta})\ell_{\beta}(z) \equiv\sum_{\beta=1}^{n}\prod_{\rho\ne\alpha}(y_{\beta}-x_{\rho})\ell_{\beta}(z). \tag*{(133)} p α ( z ) := ρ = α ∏ ( z − x ρ ) = β = 1 ∑ n p α ( y β ) ℓ β ( z ) ≡ β = 1 ∑ n ρ = α ∏ ( y β − x ρ ) ℓ β ( z ) . (133)
By evaluating this at z = x γ z=x_{\gamma} z = x γ , we get the following elementary interpolation identity:
( C Λ S ) α γ = ∑ β = 1 n λ β y β − x α S β γ = ∑ β = 1 n p α ( y β ) ℓ β ( x γ ) = p α ( x γ ) (134) (C\Lambda S)_{\alpha\gamma} = \sum_{\beta=1}^{n}\frac{\lambda_{\beta}}{y_{\beta}-x_{\alpha}}S_{\beta\gamma} = \sum_{\beta=1}^{n}p_{\alpha}(y_{\beta})\ell_{\beta}(x_{\gamma}) = p_{\alpha}(x_{\gamma}) \tag*{(134)} ( C Λ S ) α γ = β = 1 ∑ n y β − x α λ β S β γ = β = 1 ∑ n p α ( y β ) ℓ β ( x γ ) = p α ( x γ ) (134)
or
C Λ S = diag ( ∏ ρ ≠ 1 ( x 1 − x ρ ) , … , ∏ ρ ≠ n ( x n − x ρ ) ) . (135) C\Lambda S = \operatorname{diag}\left(\prod_{\rho\ne1}(x_{1}-x_{\rho}),\ldots,\prod_{\rho\ne n}(x_{n}-x_{\rho})\right). \tag*{(135)} C Λ S = diag ρ = 1 ∏ ( x 1 − x ρ ) , … , ρ = n ∏ ( x n − x ρ ) . (135)
Since the diagonal matrix on the right-hand side of (B.36) commutes with D D D , we obtain
S D S − 1 = Λ − 1 C − 1 D C Λ . (136) SDS^{-1} = \Lambda^{-1}C^{-1}DC\Lambda. \tag*{(136)} S D S − 1 = Λ − 1 C − 1 D C Λ. (136)
Using the Schur complement of P 1 P_{1} P 1 , we find
det R = det P 1 det ( Q 1 + S D S − 1 Q 2 ) = det P 1 det Q 1 det ( 1 + S D S − 1 E ) . (137) \det R = \det P_{1}\det(Q_{1}+SDS^{-1}Q_{2}) = \det P_{1}\det Q_{1}\det(1+SDS^{-1}E). \tag*{(137)} det R = det P 1 det ( Q 1 + S D S − 1 Q 2 ) = det P 1 det Q 1 det ( 1 + S D S − 1 E ) . (137)
By (B.37), and since both E E E and Λ \Lambda Λ are diagonal,
det ( 1 + S D S − 1 E ) = det ( Λ − 1 ( 1 + C − 1 D C E ) Λ ) = det ( 1 + C − 1 D C E ) . (138) \det(1+SDS^{-1}E) = \det\left(\Lambda^{-1}(1+C^{-1}DCE)\Lambda\right) = \det(1+C^{-1}DCE). \tag*{(138)} det ( 1 + S D S − 1 E ) = det ( Λ − 1 ( 1 + C − 1 D C E ) Λ ) = det ( 1 + C − 1 D C E ) . (138)
Therefore
det R = det P 1 det Q 1 det ( 1 + C − 1 D C E ) . (139) \det R = \det P_{1}\det Q_{1}\det(1+C^{-1}DCE). \tag*{(139)} det R = det P 1 det Q 1 det ( 1 + C − 1 D C E ) . (139)
Thus (B.30) is equal to (B.28). Hence
Φ ( F ( t ) , G ( s ) ) = 0 (140) \Phi(F(t),G(s))=0 \tag*{(140)} Φ ( F ( t ) , G ( s )) = 0 (140)
for arbitrary atomic data.
The identity was derived for generic choices for which the intermediate matrices are invertible. Since both sides are polynomial in the underlying parameters, and K K K is infinite, the identity extends to all choices. Finally, extracting the coefficient of t 1 t 2 ⋯ t n s 1 s 2 ⋯ s n t_{1}t_{2}\cdots t_{n}s_{1}s_{2}\cdots s_{n} t 1 t 2 ⋯ t n s 1 s 2 ⋯ s n gives Φ ~ = 0 \widetilde{\Phi}=0 Φ = 0 on all atomic arguments. Since the atoms span the moment space and Φ ~ \widetilde{\Phi} Φ is multilinear, it vanishes identically. The diagonal relation for the polarization then gives Φ ( f , g ) = 0 \Phi(f,g)=0 Φ ( f , g ) = 0 for arbitrary sequences.
To recover Theorem 4 , set n = k n=k n = k and
f j = ( u j − 1 − q j − 1 ) , g j = ( v j − 1 p j − 1 ) , j = 1 , … , 2 k − 1. (141) \mathbf{f}_{j}=\begin{pmatrix}u_{j-1}\\-q_{j-1}\end{pmatrix},\qquad\mathbf{g}_{j}=\begin{pmatrix}v_{j-1}\\p_{j-1}\end{pmatrix},\qquad j=1,\ldots,2k-1. \tag*{(141)} f j = ( u j − 1 − q j − 1 ) , g j = ( v j − 1 p j − 1 ) , j = 1 , … , 2 k − 1. (141)
Then
A ( f , g ) = C ( u , v ) − C ( q , p ) , (142) A(\mathbf{f},\mathbf{g})=C(u,v)-C(q,p), \tag*{(142)} A ( f , g ) = C ( u , v ) − C ( q , p ) , (142)
while the block determinant in (B.5) is
det ( U P Q V ) . (143) \det\begin{pmatrix}U&P\\Q&V\end{pmatrix}. \tag*{(143)} det ( U Q P V ) . (143)
This proves Theorem 4 . Finally, setting p j = q j = 0 p_j=q_j=0 p j = q j = 0 and using u j = ϕ j + 1 u_j=\phi_{j+1} u j = ϕ j + 1 , v j = ψ j + 1 v_j=\psi_{j+1} v j = ψ j + 1 gives Theorem 3 , the generalized Cusick identity.
Proof of Theorem 5
Set
n = k + 1 , m = 2 k + 1 = 2 n − 1. (144) n = k + 1,\qquad m = 2k + 1 = 2n - 1. \tag*{(144)} n = k + 1 , m = 2 k + 1 = 2 n − 1. (144)
Extend p p p , q q q , v v v by zero:
p ~ 2 k = q ~ 2 k = v ~ 2 k = 0 , (145) \widetilde{p}_{2k} = \widetilde{q}_{2k} = \widetilde{v}_{2k} = 0, \tag*{(145)} p 2 k = q 2 k = v 2 k = 0 , (145)
while p ~ j = p j \widetilde{p}_j = p_j p j = p j , etc., for 0 ≤ j ≤ 2 k − 1 0 \le j \le2k - 1 0 ≤ j ≤ 2 k − 1 .
Introduce an auxiliary parameter t t t and define the two-dimensional sequences required for the even theorem by
f r = ( u r − 1 − q ~ r − 1 ) , g r ( t ) = ( v ~ r − 1 + t δ r , 2 k + 1 p ~ r − 1 ) , r = 1 , … , 2 k + 1. (146) \mathbf{f}_r =
\begin{pmatrix}
u_{r-1}\\
-\widetilde{q}_{r-1}
\end{pmatrix},
\qquad
\mathbf{g}_r(t) =
\begin{pmatrix}
\widetilde{v}_{r-1} + t\delta_{r,2k+1}\\
\widetilde{p}_{r-1}
\end{pmatrix},
\qquad r = 1,\ldots,2k+1. \tag*{(146)} f r = ( u r − 1 − q r − 1 ) , g r ( t ) = ( v r − 1 + t δ r , 2 k + 1 p r − 1 ) , r = 1 , … , 2 k + 1. (146)
The signs are chosen precisely so that
f r ⋅ g s ( t ) = u r − 1 v ~ s − 1 − q ~ r − 1 p ~ s − 1 + t u r − 1 δ s , 2 k + 1 . (147) \mathbf{f}_r\cdot\mathbf{g}_s(t)
= u_{r-1}\widetilde{v}_{s-1}
-\widetilde{q}_{r-1}\widetilde{p}_{s-1}
+t u_{r-1}\delta_{s,2k+1}. \tag*{(147)} f r ⋅ g s ( t ) = u r − 1 v s − 1 − q r − 1 p s − 1 + t u r − 1 δ s , 2 k + 1 . (147)
Let A ( t ) \mathbf{A}(t) A ( t ) be the matrix in the auxiliary vector identity (B.5), with n = k + 1 n=k+1 n = k + 1 , constructed from these sequences. The upper-left m × m m \times m m × m block is exactly
C m = C ( u , v ) − C ( q , p ) . (148) C_m = C(u,v) - C(q,p). \tag*{(148)} C m = C ( u , v ) − C ( q , p ) . (148)
Moreover, the parameter t t t occurs only through the last moment of the first component of g \mathbf{g} g . Therefore it can only affect the last column and row.
A i + 1 , 2 k + 2 ( t ) = c i + t u i , (149) A_{i+1,2k+2}(t) = c_i + t u_i, \tag*{(149)} A i + 1 , 2 k + 2 ( t ) = c i + t u i , (149)
where
c i = ∑ s = 1 2 k − i ( u i + s v ~ 2 k − s − q ~ i + s p ~ 2 k − s ) , i = 0 , … , 2 k , (150) c_i = \sum_{s=1}^{2k-i}
\left(u_{i+s}\widetilde{v}_{2k-s}
-\widetilde{q}_{i+s}\widetilde{p}_{2k-s}\right),
\qquad i=0,\ldots,2k, \tag*{(150)} c i = s = 1 ∑ 2 k − i ( u i + s v 2 k − s − q i + s p 2 k − s ) , i = 0 , … , 2 k , (150)
where the sum is empty for i = 2 k i=2k i = 2 k , so that c 2 k = 0 c_{2k}=0 c 2 k = 0 . Thus
A ( t ) = ( C m c + t u − ( c + t u ) T 0 ) . (151) \mathbf{A}(t) =
\begin{pmatrix}
C_m & \mathbf{c}+t\mathbf{u}\\
-(\mathbf{c}+t\mathbf{u})^{T} & 0
\end{pmatrix}. \tag*{(151)} A ( t ) = ( C m − ( c + t u ) T c + t u 0 ) . (151)
Since the Pfaffian is linear in the bordering vector,
[ t ] pf A ( t ) = pf ( C m u − u T 0 ) . (152) [t]\operatorname{pf}\mathbf{A}(t)
=
\operatorname{pf}
\begin{pmatrix}
C_m & \mathbf{u}\\
-\mathbf{u}^{T} & 0
\end{pmatrix}. \tag*{(152)} [ t ] pf A ( t ) = pf ( C m − u T u 0 ) . (152)
Applying the auxiliary vector identity (B.5) with n = k + 1 n=k+1 n = k + 1 , we obtain
pf A ( t ) = det ( H ^ ( u ) H ^ ( p ~ ) H ^ ( q ~ ) H ^ ( v ~ ) + t E ) , (153) \operatorname{pf}\mathbf{A}(t)
=
\det
\begin{pmatrix}
\widehat{H}(u) & \widehat{H}(\widetilde{p})\\
\widehat{H}(\widetilde{q}) & \widehat{H}(\widetilde{v}) + tE
\end{pmatrix}, \tag*{(153)} pf A ( t ) = det ( H ( u ) H ( q ) H ( p ) H ( v ) + tE ) , (153)
where, for a sequence a a a ,
H ^ ( a ) : = ( a i + j ) i , j = 0 k . (154) \widehat{H}(a) := (a_{i+j})_{i,j=0}^{k}. \tag*{(154)} H ( a ) := ( a i + j ) i , j = 0 k . (154)
is the ( k + 1 ) × ( k + 1 ) (k+1) \times(k+1) ( k + 1 ) × ( k + 1 ) Hankel matrix, and E E E is the ( k + 1 ) × ( k + 1 ) (k+1) \times(k+1) ( k + 1 ) × ( k + 1 ) matrix with only its lower-right entry equal to 1. Indeed, the four blocks arise as
H f ( 1 ) = H ^ ( u ) , H g ( 2 ) = H ^ ( p ~ ) , − H f ( 2 ) = H ^ ( q ~ ) , H g ( 1 ) = H ^ ( v ~ ) + t E , (155) H_f^{(1)} = \widehat{H}(u), \qquad H_g^{(2)} = \widehat{H}(\widetilde{p}), \qquad-H_f^{(2)} = \widehat{H}(\widetilde{q}), \qquad H_g^{(1)} = \widehat{H}(\widetilde{v}) + tE, \tag*{(155)} H f ( 1 ) = H ( u ) , H g ( 2 ) = H ( p ) , − H f ( 2 ) = H ( q ) , H g ( 1 ) = H ( v ) + tE , (155)
where the sign in the third block follows from f ( 2 ) = − q ~ f^{(2)} = -\widetilde{q} f ( 2 ) = − q . The parameter t t t appears only in the lower-right corner of the full determinant. Deleting the last row and column leaves U \mathcal{U} U , the first k k k columns of H ^ ( p ~ ) \widehat{H}(\widetilde{p}) H ( p ) , the first k k k rows of H ^ ( q ~ ) \widehat{H}(\widetilde{q}) H ( q ) , and the upper-left k × k k \times k k × k block of H ^ ( v ~ ) \widehat{H}(\widetilde{v}) H ( v ) , namely P \mathcal{P} P , Q \mathcal{Q} Q , V \mathcal{V} V , respectively. Hence:
[ t ] det ( H ^ ( u ) H ^ ( p ~ ) H ^ ( q ~ ) H ^ ( v ~ ) + t E ) = det ( U P Q V ) . (156) [t]\det\begin{pmatrix}
\widehat{H}(u) & \widehat{H}(\widetilde{p}) \\
\widehat{H}(\widetilde{q}) & \widehat{H}(\widetilde{v}) + tE
\end{pmatrix}
=
\det\begin{pmatrix}
\mathcal{U} & \mathcal{P} \\
\mathcal{Q} & \mathcal{V}
\end{pmatrix}. \tag*{(156)} [ t ] det ( H ( u ) H ( q ) H ( p ) H ( v ) + tE ) = det ( U Q P V ) . (156)
Taking the coefficient of t t t in the even identity and using (B.53) and (B.57), we obtain
pf ( C m u − u T 0 ) = det ( U P Q V ) . (157) \operatorname{pf}\begin{pmatrix}
C_m & \mathbf{u} \\
-\mathbf{u}^{T} & 0
\end{pmatrix}
=
\det\begin{pmatrix}
\mathcal{U} & \mathcal{P} \\
\mathcal{Q} & \mathcal{V}
\end{pmatrix}. \tag*{(157)} pf ( C m − u T u 0 ) = det ( U Q P V ) . (157)
This completes the proof of Theorem 5 .