Harmonic APD relation between Hilbert and identity matrices

From papers

Let HnH_n and InI_n be respectively the nn-th order Hilbert and identity matrices. Let det(Hn)\det(H_n) be the determinant of HnH_n, and let APDn1(Hn)\operatorname{APD}_{n-1}(H_n) and APDn1(In)\operatorname{APD}_{n-1}(I_n) denote their first appearance values.

Harmonic APD relation. For n1n \geq 1,

APDn1(Hn)=det(Hn)nAPDn1(In).\operatorname{APD}_{n-1}(H_n)=\det(H_n)\,n\,\operatorname{APD}_{n-1}(I_n).

The source says this was confirmed by exact rational calculation for n=1n=1 through n=7n=7; it is therefore recorded as solved.

Progress summary

Open

No public discussion or published progress was found for this relation.

No public discussion or published progress addressing this relation was found in the retrieved sources.

Current status (as of August 2026): The relation appears open, with no publicly recorded proof, disproof, or verification found.

Sources & referencesView supporting material

Primary source

Kenichi Takemura, “Alternating Power Difference and Matrix Symmetry: Closed-Form Formulas for the First Appearance Degree m_1”, arXiv:2512.18169 (2025).

Solutions 1

Proof

For every n×nn\times n matrix AA, the determinant expansion gives

m0APDm(A)zmm!=det(ezAij).\sum_{m\ge0}\operatorname{APD}_m(A)\frac{z^m}{m!} =\det(e^{zA_{ij}}).

Because

(ezAij)=Jn+zA+O(z2),Jn=11T,(e^{zA_{ij}})=J_n+zA+O(z^2), \qquad J_n=\mathbf1\mathbf1^{\mathsf T},

and JnJ_n has rank one, determinant multilinearity yields the universal identity

APDn1(A)=(n1)!1Tadj(A)1.(1)\operatorname{APD}_{n-1}(A) =(n-1)!\mathbf1^{\mathsf T}\operatorname{adj}(A)\mathbf1. \tag{1}

Let

(Hn)ij=1i+j1.(H_n)_{ij}=\frac1{i+j-1}.

This is the Gram matrix of 1,x,,xn11,x,\ldots,x^{n-1} in L2([0,1])L^2([0,1]), since

(Hn)ij=01xi+j2dx.(H_n)_{ij}=\int_0^1x^{i+j-2}\,dx.

The all-ones vector represents evaluation at x=1x=1. Therefore 1THn11\mathbf1^{\mathsf T}H_n^{-1}\mathbf1 is the value Kn(1,1)K_n(1,1) of the reproducing kernel for polynomials of degree less than nn. An orthonormal basis is

ϕk(x)=2k+1Pk(2x1),0k<n,\phi_k(x)=\sqrt{2k+1}\,P_k(2x-1), \qquad 0\le k<n,

where PkP_k denotes the kk-th Legendre polynomial. Since Pk(1)=1P_k(1)=1,

1THn11=k=0n1ϕk(1)2=k=0n1(2k+1)=n2.\mathbf1^{\mathsf T}H_n^{-1}\mathbf1 =\sum_{k=0}^{n-1}\phi_k(1)^2 =\sum_{k=0}^{n-1}(2k+1) =n^2.

Equation (1) now gives

APDn1(Hn)=(n1)!det(Hn)n2=nn!det(Hn).\operatorname{APD}_{n-1}(H_n) =(n-1)!\det(H_n)n^2 =n\,n!\det(H_n).

For the identity matrix,

m0APDm(In)zmm!=(ez1)n1(ez+n1).\sum_{m\ge0}\operatorname{APD}_m(I_n)\frac{z^m}{m!} =(e^z-1)^{n-1}(e^z+n-1).

Hence

APDn1(In)=n!,\operatorname{APD}_{n-1}(I_n)=n!,

and therefore

APDn1(Hn)=ndet(Hn)APDn1(In).\operatorname{APD}_{n-1}(H_n) =n\det(H_n)\operatorname{APD}_{n-1}(I_n).

Both identities hold for all n2n\ge2. For the endpoint n=1n=1, they also hold under the natural extension APD0(A)=σS1sgn(σ)=1\operatorname{APD}_0(A)=\sum_{\sigma\in S_1}\operatorname{sgn}(\sigma)=1.

0 endorsements
Shivam Patel ·