Multiplier sequences
A real sequence is a multiplier sequence if the diagonal operator it defines preserves real-rootedness; these are classified by Jensen polynomials and by the Laguerre–Pólya class.
Results
A polynomial is real-rooted if it is zero or splits over $\mathbb R$, and PF (Pólya frequency) if it is zero or has nonnegative coefficients and only real, nonpositive roots. For a real sequence $\gamma=(\gamma_0,\gamma_1,\dotsc)$ the diagonal operator is $$ a_0+a_1x+\dotsb+a_nx^n\ \mapsto\ \gamma_0a_0+\gamma_1a_1x+\dotsb+\gamma_na_nx^n . $$ The sequence $\gamma$ is a multiplier sequence if this operator maps every real-rooted polynomial to a real-rooted polynomial, and a PF multiplier sequence if it maps every PF polynomial to a PF polynomial. The finite versions restrict to inputs of degree at most $n$. The Jensen polynomial of $\gamma$ is $g_n(x)=\sum_k\binom nk\gamma_kx^k$.
Theorem (Pólya–Schur). Let $\gamma_k\ge0$ for all $k$. Then $\gamma$ is a multiplier sequence if and only if every Jensen polynomial $g_n$ is PF.
Theorem (Laguerre–Pólya classification). Assume $\sum_k|\gamma_k|R^k/k!<\infty$ for some $R>0$, and let $f(z)=\sum_k\gamma_kz^k/k!$.
- $\gamma$ is a PF multiplier sequence if and only if $f$ is in the Laguerre–Pólya class of type I, that is, a locally uniform limit of PF polynomials.
- $\gamma$ is a multiplier sequence if and only if one of $f(z)$, $-f(z)$, $f(-z)$, $-f(-z)$ is of type I.
Every multiplier sequence $\gamma$ satisfies one of the following: $\gamma$, $-\gamma$, $\bigl((-1)^k\gamma_k\bigr)$ or $\bigl(-(-1)^k\gamma_k\bigr)$ is a PF multiplier sequence.
Theorem (log-concavity). If $\gamma$ is a PF multiplier sequence, then $\gamma_k\gamma_{k+2}\le\gamma_{k+1}^2$ for all $k$.
Theorem (Laguerre). Let $\varphi$ be a real-rooted polynomial with only nonpositive roots. Then $\varphi(0),\varphi(1),\varphi(2),\dotsc$ is a multiplier sequence, and a PF multiplier sequence if $\varphi$ has nonnegative leading coefficient. In particular $k+r$ is a PF multiplier sequence for every real $r\ge0$. More generally, if $f$ is of type I, then $\bigl(\operatorname{Re}f(k)\bigr)_{k\ge0}$ is a PF multiplier sequence.
Theorem (examples). For real $\alpha>0$, the sequence $1/(\alpha)_k$, with $(\alpha)_k=\alpha(\alpha+1)\dotsm(\alpha+k-1)$, is a PF multiplier sequence. In particular $1/k!$ is one.
Theorem (operators). If $\varphi$ is Laguerre–Pólya and $g$ is real-rooted, then $\varphi(D)\,g=\sum_k\operatorname{Re}\bigl(\varphi^{(k)}(0)/k!\bigr)\,g^{(k)}$ is real-rooted. In particular, for $a\ge0$ the operator $e^{-aD^2}$ maps real-rooted polynomials to real-rooted polynomials.
Theorem (interlacing). If $\gamma$ is a multiplier sequence with $\gamma_k>0$ for all $k$ and $f\ll g$, then $\sum_k\gamma_kf_kx^k\ll\sum_k\gamma_kg_kx^k$.
Theorem (bidiagonal preservers). For sequences $\alpha,\beta$, the bidiagonal operator sends $\sum_ka_kx^k$ to $\sum_k(\alpha_ka_k+\beta_{k-1}a_{k-1})x^k$, with the second term omitted for $k=0$. It maps PF polynomials of degree at most $d$ to PF polynomials in each of the following cases:
- its degree-$d$ Borcea–Brändén algebraic symbol is stable, and $\alpha_k,\beta_k\ge0$ for $k\le d$;
- with $J_\alpha=\sum_k\binom dk\alpha_kx^k$ and $J_\beta$ defined likewise, the polynomials $J_\alpha$, $xJ_\beta$ and $J_\alpha+\lambda xJ_\beta$ are PF for every $\lambda\ge0$.
Theorem (Jensen-pencil contraction). Call $(A,B)$ compatible if every nonnegative linear combination of $A$ and $B$ is zero or real-rooted. Let $*_d$ be Schur–Szegő composition (see the Hadamard products page). Let $A$, $xB$ and $p$ be PF of degree at most $d$ with $\deg B\le d$, and suppose $(A,xB)$ is compatible. Then $(A*_dp,\ x\,(B*_dp))$ is compatible. Two steps of the proof hold more generally.
- If $A,B\ne0$ are PF, $(A,xB)$ is compatible and $\varepsilon>0$, then $\bigl(A(x+\varepsilon),\ x\,B(x+\varepsilon)\bigr)$ is compatible.
- A coefficientwise continuous curve of PF sequences has a PF endpoint.
Theorem (Bernstein-diagonal preservers). Fix $D\ge1$, let $B_k=\binom Dk x^k(1+x)^{D-k}$, let $\gamma$ be a real sequence, let $L$ be the linear map with $LB_k=\gamma_kB_k$, and put $g=L\bigl((1+2x)^D\bigr)=\sum_k\gamma_kB_k$. Let $\mathcal C_D$ be the set of polynomials of degree exactly $D$ with positive leading coefficient and all roots in $[-1,0]$. Then $$ L(\mathcal C_D)\subseteq\mathcal C_D\cup\{0\}\iff g\in\mathcal C_D\cup\{0\}. $$ If these hold, then $Lf\ne0$ for all $f\in\mathcal C_D$ if and only if $g$ is real-rooted with all roots in $(-1,0)$.
Proof
The Bernstein-diagonal map is conjugate by the Cayley change of variable $J_Df(z)=(1-z)^Df\bigl(z/(1-z)\bigr)$ to the diagonal operator of $\gamma$, which turns $B_k$ into $\binom Dkz^k$ and $L$ into Schur–Szegő composition with $J_Dg$, so the statement follows from [KS06, Prop. 1.2 and Thm. 1.3].
References
- [SP14] Über zwei Arten von Faktorenfolgen in der Theorie der algebraischen Gleichungen. J. Reine Angew. Math., 1914.
BibTeX
@article{SchurPolya1914, title={{\"U}ber zwei {A}rten von {F}aktorenfolgen in der {T}heorie der algebraischen {G}leichungen}, author={Schur, J and Polya, G}, year={1914}, journal={J. Reine Angew. Math.}, publisher={Walter de Gruyter, Berlin/New York Berlin, New York} } - [KS06] On the Schur--Szegö composition of polynomials. Comptes Rendus. Mathématique, 343(2), 81-86, 2006.
BibTeX
@article{KostovShapiro2006, author = {Kostov, Vladimir and Shapiro, Boris}, title = {On the {S}chur--{S}zeg{\"o} composition of polynomials}, year = {2006}, journal = {Comptes Rendus. Mathématique}, volume = {343}, number = {2}, pages = {81--86}, doi = {10.1016/j.crma.2006.06.007}, url = {https://doi.org/10.1016/j.crma.2006.06.007} }