Search arXivSearch

arXiv subjects

Eric Hou

Publications and source records attributed to Eric Hou.

3 recordsLinked to original sources

Cofinite Zeros of High Derivatives

We give a bounded-coefficient probabilistic construction of a transcendental entire function $f$ of order two such that every nonempty open subset of the complex plane contains a zero of $f^{(n)}$ for all sufficiently large $n$. This gives an affirmative answer to the transcendental form of Erdős Problem~906. Thus every fixed disk is zero-free for only finitely many successive derivatives. The function satisfies $|f(z)|\leq\sqrt2\exp(|z|^2)$ and is a counterexample to a theorem of Boas and Reddy as printed.

math.CV

Coprime Actions and Characters Non-vanishing on the Fixed-point Subgroup

Let a finite group $A$ act coprimely on a finite group $G$ and put $C=C_G(A)$. A classical theorem of Burnside asserts that the irreducible characters of a group vanishing nowhere are exactly the linear ones, and Navarro asked in Problem 21.100 of the 21st Kourovka Notebook whether the coprime analogue holds: is the number of $A$-invariant $χ\in\Irr(G)$ with $χ_C$ nowhere zero always $|C/C'|$? We answer this negatively. For $A$ cyclic of order $21$ acting on $G=B\rtimes V$, where $V=\F_4^2\oplus\F_8$ and $B$ is the group of Boolean functions on $V$, the fixed subgroup $C\cong C_2^{12}$ carries all $4096$ invariant characters while only $1728$ of them are nowhere zero on $C$. Because $C$ is abelian the same example refutes Problem 6.3 of Navarro's problem list, and with it the corresponding statement about the head characters of Isaacs; passing to $G\rtimes A$ refutes Problem 6.7, a conjecture Isaacs reports is supported by abundant computational evidence. The construction needs only that $V$ have an $A$-stable subset of half its size. This holds for infinitely many pairs $(A,V)$, and for none of dimension below $7$, so the example is minimal over all operator groups of odd order. Exact machine verifications, independent of the proofs, accompany the paper.

math.GR

Local Regularization Does Not Characterize Multiclass PAC Learnability

Local regularization assigns each hypothesis a test-point-dependent score and predicts with a minimum-score hypothesis consistent with the sample. Asilis et al. asked whether this principle characterizes multiclass PAC learnability. We give a negative answer. There is a countable class of Daniely--Shalev-Shwartz dimension at most two with realizable PAC sample complexity \[ O\!\left(\frac{1}{\varepsilon}\log\frac{1}δ\right), \] that no local regularizer learns. Hypotheses are edges of complete graphs and instances are tournaments. At a test tournament, the scores fix an edge ranking while the training sample independently removes competitors. Cyclic triangles force enough inversions that surviving competitors produce constant population error at arbitrarily large sample sizes.

cs.LG