Skip to content
DraconiticResearch
Search Sources API
197,688 studies31 sourcesUpdated just now
PreprintarXiv licencearXiv

On a positivity conjecture of Fasi and Higham for Padé approximants to the matrix exponential

Abstract

We study the positivity conjecture of Fasi and Higham [SIAM J. Matrix Anal. Appl. 40 (2019), Conjecture 3.2], which concerns the Taylor coefficients of the reciprocal of the denominator qk,m(z) of the Padé approximant to the matrix exponential. The conjecture asserts that these coefficients are positive whenever m is odd. We prove full positivity for k=0 and eventual positivity whenever the unique positive real zero of qk,m is the unique zero of minimal modulus. We also give an explicit computable sufficient threshold beyond which all coefficients are positive. Using Hermite scaling with leading-coefficient normalization, we prove that this spectral minimality holds for each fixed odd m and all sufficiently large k. The unrestricted conjecture is nevertheless false. We give exact rational counterexamples and prove a transient scaling theorem showing that on the scale nasymp√(k+1) the normalized coefficients converge to an explicit oscillatory Hermite profile. For m=5 this profile is negative at one point, yielding infinitely many counterexamples. The counterexamples are not diagonal, and the diagonal case k=m, which is the case relevant in practice, remains open. These results do not imply a failure of the Fasi--Higham algorithm. Rather, they delimit the range in which the auxiliary positivity conjecture can be used as a theorem.

Related

  1. On the exponential type conjecture
    Zihong Chen
    Journal of the American Mathematical SocietyPeer-reviewedLicence not statedFull text12 citedGeometry & topologyOpenAlex
  2. BMV conjecture and positivity of spectral shift for relatively Schatten perturbations
    Chandan Pradhan, Anna Skripka
    arXivPreprintarXiv licenceFull textAnalysis & PDEsMathematical physicsarXiv
  3. A positivity conjecture of Prellberg and Stanton
    Thomas Prellberg
    arXivPreprintCC BY 4.0Full textCombinatorics & discrete mathematicsarXiv
  4. Linear bounds and sharp exponential growth for commuting matrix algebra
    Jiamei Li, Caihua Luo, Xu Song +1
    arXivPreprintCC0Full textAlgebra & number theoryarXiv
  5. An Exponential Lower Bound for the Permanent of Random Bernoulli Matrix
    Yiming Chen
    arXivPreprintarXiv licenceFull textProbabilityarXiv
BibTeX
@misc{scardino2026positivity,
  title = {{On a positivity conjecture of Fasi and Higham for Padé approximants to the matrix exponential}},
  author = {Francesco Scardino},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.39051},
  eprint = {2610.39051},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.39051},
}
RIS
TY  - UNPB
TI  - On a positivity conjecture of Fasi and Higham for Padé approximants to the matrix exponential
AU  - Francesco Scardino
PY  - 2026
DA  - 2026/10/28
PB  - arXiv
DO  - 10.48550/arxiv.2610.39051
UR  - https://arxiv.org/abs/2610.39051
AB  - We study the positivity conjecture of Fasi and Higham [SIAM J. Matrix Anal. Appl. 40 (2019), Conjecture 3.2], which concerns the Taylor coefficients of the reciprocal of the denominator q_k,m(z) of the Padé approximant to the matrix exponential. The conjecture asserts that these coefficients are positive whenever m is odd. We prove full positivity for k=0 and eventual positivity whenever the unique positive real zero of q_k,m is the unique zero of minimal modulus. We also give an explicit computable sufficient threshold beyond which all coefficients are positive. Using Hermite scaling with leading-coefficient normalization, we prove that this spectral minimality holds for each fixed odd m and all sufficiently large k. The unrestricted conjecture is nevertheless false. We give exact rational counterexamples and prove a transient scaling theorem showing that on the scale nasymp√(k+1) the normalized coefficients converge to an explicit oscillatory Hermite profile. For m=5 this profile is negative at one point, yielding infinitely many counterexamples. The counterexamples are not diagonal, and the diagonal case k=m, which is the case relevant in practice, remains open. These results do not imply a failure of the Fasi--Higham algorithm. Rather, they delimit the range in which the auxiliary positivity conjecture can be used as a theorem.
ER  -