← All problems

GLT geometric-mean conjecture

Let {An}n\{A_n\}_n and {Bn}n\{B_n\}_n be sequences of Hermitian positive-definite matrices belonging to the GLT class, with scalar GLT symbols κ\kappa and ξ\xi, respectively. If κ(x,θ)ξ(x,θ)=ξ(x,θ)κ(x,θ)\kappa(x,\theta)\xi(x,\theta)=\xi(x,\theta)\kappa(x,\theta) almost everywhere, then, without assuming that either symbol is invertible almost everywhere, the matrix geometric-mean sequence {G(An,Bn)}n\{G(A_n,B_n)\}_n is GLT and satisfies {G(An,Bn)}n∼GLT(κξ)1/2\{G(A_n,B_n)\}_n\sim_{\mathrm{GLT}}(\kappa\xi)^{1/2}, where G(An,Bn)=An1/2(An−1/2BnAn−1/2)1/2An1/2G(A_n,B_n)=A_n^{1/2}(A_n^{-1/2}B_nA_n^{-1/2})^{1/2}A_n^{1/2} and (κξ)1/2(\kappa\xi)^{1/2} denotes the pointwise positive square root.

References

Progress summary

Refreshed
Open · solution reported

A preprint claims to settle the conjecture for commuting matrix symbols and to show that the degenerate noncommuting case generally fails, but the claim has not been independently checked.

The GLT geometric-mean conjecture concerns the spectral symbol of the matrix geometric mean of two positive-definite GLT sequences. The claimed result removes the earlier invertibility requirement when the symbols commute and addresses degenerate noncommuting symbols.

Known results

  • A January 24, 2025 paper proved the geometric-mean formula assuming at least one symbol is invertible almost everywhere, and left removal of that assumption under commutativity as future work.
  • Ilyas, Khan, Loi, and Serra-Capizzano (2025) claimed that commuting symbols suffice: {G(An,Bn)}n∼GLT(κξ)1/2\{G(A_n,B_n)\}_n\sim_{\mathrm{GLT}}(\kappa\xi)^{1/2}, including r=1r=1 and d≥1d\ge 1.

New thesis-like preprint report (2026)

A newly posted thesis-like preprint by Asim Ilyas presents the commuting-symbol theorem as settling the conjecture in the scalar-block setting and treats degenerate noncommuting cases, where the formula may be false or not well posed. The result is a claimed resolution, not an independently verified one.

Current status (as of October 2026): The commuting, non-invertible case is claimed solved, while the broader noncommuting degenerate formulation is reported to fail or be ill posed; the claims remain unverified.

Sources

Solutions 0

No solutions have been posted yet.