Roverato's conjecture on G-Wishart normalising constants

From papers

Let G\mathcal G be a graph with nn vertices. For DS++nD\in\mathbb S^n_{++}, let DGD^{\mathcal G} denote the positive-definite completion of DD with respect to G\mathcal G, and let IssG(DG)\operatorname{Iss}_{\mathcal G}(D^{\mathcal G}) be the corresponding Isserlis matrix. Write CG(δ,D)\mathscr C_{\mathcal G}(\delta,D) for the G\mathcal G-Wishart normalising constant and let InI_n be the n×nn\times n identity matrix.

Roverato's conjecture. For every real number δ>0\delta>0,

CG(δ,D)=2n2det ⁣(IssG(DG))12det(DG)δ22CG(δ,In).\mathscr C_{\mathcal G}(\delta,D)=2^{\frac n2}\det\!\left(\operatorname{Iss}_{\mathcal G}(D^{\mathcal G})\right)^{-\frac12}\det(D^{\mathcal G})^{-\frac{\delta-2}{2}}\mathscr C_{\mathcal G}(\delta,I_n).

The conjecture would reduce the evaluation of the normalising constant for a general scale matrix to the identity-scale case, with the remaining factors obtainable from positive-definite completion and determinant calculations. It is known for chordal graphs, but the source presents it as unresolved in general.

Progress summary

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Sources & referencesView supporting material

Primary source

Ching Wong, Giusi Moffa and Jack Kuipers, “On a conjecture of Roverato regarding G-Wishart normalising constants”, arXiv:2503.13046 (2025).

Solutions 0

No solutions have been posted yet.