bibkey: godsil2013averagemixing authors: Chris Godsil year: 2013 title: “Average mixing of continuous quantum walks” doi: 10.1016/j.jcta.2013.05.006 url: https://arxiv.org/abs/1103.2578v3 claim: “Section 11, Question 1 asks whether the discriminant D of the adjacency minimal polynomial always makes D times the average mixing matrix integral. Lemma 1.1 gives the spectral Schur-square formula; Lemma 3.1 proves the D squared bound.” strata_touched:
- D5/S3/Quantum/Dynamics/AverageMixingDiscriminantIntegrality license: citation-only triage: anchor
Average mixing of continuous quantum walks
Chris Godsil, arXiv:1103.2578v3; Journal of Combinatorial Theory, Series A 120 (2013), 1649–1662, doi:10.1016/j.jcta.2013.05.006. Page numbers below refer to the arXiv v3 PDF.
Section 11, Question 1 (page 20) asks:
Is it true that if is the discriminant of the minimal polynomial of , then is an integer matrix?
The minimal polynomial of the graph means the minimal polynomial of its adjacency matrix over the rationals. Section 1 (page 2) specifies the spectral projections:
where runs over the distinct eigenvalues of and is the matrix representing orthogonal projection onto the eigenspace belonging to .
Lemma 1.1 (page 3) states:
If is the spectral decomposition of , then .
Here is the Schur (entrywise) product and .
This is the spectral expression used by the avgMixing definition
in D5/S3/Quantum/Dynamics/AverageMixingTraceMaximum.lean.
Lemma 3.1 (page 5) states:
If is the discriminant of the minimal polynomial of , then is an integer matrix.
Question 1 asks for the stronger factor . Questions 2 and 3 of section 11 concern an algorithm over the rationals and the graphs whose average mixing matrix is a linear combination of and .
Verified locator
DOI: https://doi.org/10.1016/j.jcta.2013.05.006. Source version: https://arxiv.org/abs/1103.2578v3. The quoted statements occur on PDF pages 3 (Lemma 1.1), 5 (Lemma 3.1) and 20 (section 11, Question 1).