Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Let p=z4+az2+bz+cp=z^4+az^2+bz+c satisfy the claim. Its three critical points are simple, since it has three distinct critical values. Polynomial division gives

Vp(z)=z54+3az38+7bz216+(a216+c2)z+3ab16,Qp(z)=Vp′(z)=54z4+9a8z2+7b8z+a216+c2,Rp(z)=p(z)2−p′(z)Vp(z)=r2z2+r1z+r0,r2=ac−a38+9b216,r1=32bc−7a2b16,r0=c2−3ab216.(1)\begin{aligned} V_p(z)&=\frac{z^5}{4}+\frac{3a z^3}{8}+\frac{7b z^2}{16} +\left(\frac{a^2}{16}+\frac c2\right)z+\frac{3ab}{16},\\ Q_p(z)&=V_p'(z)=\frac54z^4+\frac{9a}{8}z^2+\frac{7b}{8}z +\frac{a^2}{16}+\frac c2,\\ R_p(z)&=p(z)^2-p'(z)V_p(z)=r_2z^2+r_1z+r_0,\\ r_2&=ac-\frac{a^3}{8}+\frac{9b^2}{16},\qquad r_1=\frac32bc-\frac{7a^2b}{16},\qquad r_0=c^2-\frac{3ab^2}{16}. \end{aligned} \tag{1}

A local family with prescribed critical values

Solve the holomorphic coefficient differential equation

∂τqτ=Rqτ,q0=p.(2)\partial_\tau q_\tau=R_{q_\tau},\qquad q_0=p. \tag{2}

Its polynomial right-hand side gives a unique holomorphic solution for τ\tau in a neighborhood of zero, consisting of centered monic quartics. The critical points remain simple there. At a moving critical point cj(τ)c_j(\tau) its value vj(τ)=qτ(cj(τ))v_j(\tau)=q_\tau(c_j(\tau)) satisfies

vj′(τ)=Rqτ(cj(τ))=vj(τ)2,vj(τ)=vj1−τvj.(3)v_j'(\tau)=R_{q_\tau}(c_j(\tau))=v_j(\tau)^2, \qquad v_j(\tau)=\frac{v_j}{1-\tau v_j}. \tag{3}

Shrink the parameter neighborhood to lie in ∣τ∣<1|\tau|<1 and define

pτ(z)=(1−∣τ∣2)qτ((1−∣τ∣2)−1/4z)−τˉ.(4)p_\tau(z)=(1-|\tau|^2) q_\tau\bigl((1-|\tau|^2)^{-1/4}z\bigr)-\bar\tau. \tag{4}

This is a real analytic centered monic family. Its critical values are

vj−τˉ1−τvj,(5)\frac{v_j-\bar\tau}{1-\tau v_j}, \tag{5}

which lie on the unit circle and stay distinct. No global continuation or theorem about the critical-value map is needed here.

Differentiating the length

For ζ\zeta on the unit circle and different from the three vjv_j, each solution w=w(τ,ζ)w=w(\tau,\zeta) of

qτ(w)=ζ1−τζq_\tau(w)=\frac{\zeta}{1-\tau\zeta}

is locally holomorphic in τ\tau. Differentiating and using (2) gives

∂τw=Vqτ(w).(6)\partial_\tau w=V_{q_\tau}(w). \tag{6}

On each of the open arcs between consecutive vjv_j choose the four inverse branches and let h=∂θwh=\partial_\theta w, where ζ=eiθ\zeta=e^{i\theta}. This tangent derivative is nonzero, and (6) implies

∂τlog⁡h=Qqτ(w),∂τ2log⁡h∣τ=0=Wp(w),(7)\partial_\tau\log h=Q_{q_\tau}(w),\qquad \partial_\tau^2\log h\big|_{\tau=0}=W_p(w), \tag{7}

where, by differentiating (1) in the coefficient direction RpR_p,

Wp=VpQp′+9r28z2+7r18z+ar28+r02.(8)W_p=V_pQ_p'+\frac{9r_2}{8}z^2+\frac{7r_1}{8}z +\frac{ar_2}{8}+\frac{r_0}{2}. \tag{8}

The singular endpoints cause no loss of differentiability. Near a simple critical point the inverse branches have a convergent square-root expansion. The parameterization in (3) moves the endpoint and the target by the same fractional linear map, so that the singular factor is ∣θ−θj∣−1/2|\theta-\theta_j|^{-1/2} at a fixed value of θj\theta_j. The expansion coefficients are holomorphic in τ\tau and bounded on a smaller closed parameter disc, since the critical points remain simple and their values remain distinct. Consequently ∣h∣|h| has a common integrable majorant. The factors Qqτ(w)Q_{q_\tau}(w) and their first parameter derivatives are bounded there as well, since all the roots in question stay in a compact set. Thus the first and second parameter derivatives of ∣h∣|h| have a common integrable majorant. Differentiation under the sum and integral below is justified.

The fractional linear image of the unit circle is the circle with center τˉ/(1−∣τ∣2)\bar\tau/(1-|\tau|^2) and radius (1−∣τ∣2)−1(1-|\tau|^2)^{-1}. The length of its inverse image under qτq_\tau is

L(τ)=∑arcs and branches∫∣h(τ,θ)∣ dθ,ℓ(pτ)=(1−∣τ∣2)1/4L(τ).(9)L(\tau)=\sum_{\text{arcs and branches}}\int |h(\tau,\theta)|\,d\theta, \qquad \ell(p_\tau)=(1-|\tau|^2)^{1/4}L(\tau). \tag{9}

The sum counts every regular arc of the lemniscate once; the finitely many critical points have zero arclength. If hh is nonzero and holomorphic in τ\tau, (7) gives at zero

∂τ∣h∣=12Q∣h∣,∂τ∂τˉ∣h∣=14∣Q∣2∣h∣,∂τ2∣h∣=(12W+14Q2)∣h∣.(10)\begin{aligned} \partial_\tau |h|&=\tfrac12Q|h|,\\ \partial_\tau\partial_{\bar\tau}|h|&=\tfrac14|Q|^2|h|,\\ \partial_\tau^2|h|&=(\tfrac12W+\tfrac14Q^2)|h|. \end{aligned} \tag{10}

Divide the integrals by L(0)=ℓ(p)L(0)=\ell(p) and write E\mathbb E for normalized arclength expectation on E(p)E(p). With m=EQm=\mathbb E Q, equations (9)--(10), including the derivative of its real prefactor, give for u=log⁡ℓ(pτ)u=\log\ell(p_\tau)

2uτ=m,4uττˉ=E∣Q∣2−∣m∣2−1,2uττ=EW+12(EQ2−m2).(11)\begin{aligned} 2u_\tau&=m,\\ 4u_{\tau\bar\tau}&=\mathbb E|Q|^2-|m|^2-1,\\ 2u_{\tau\tau}&=\mathbb E W+\tfrac12(\mathbb E Q^2-m^2). \end{aligned} \tag{11}

All derivatives in (11) are evaluated at zero. In particular Δτu=Var⁡(Q)−1\Delta_\tau u=\operatorname{Var}(Q)-1 and Δτℓ(pτ)/ℓ(p)=E∣Q∣2−1\Delta_\tau\ell(p_\tau)/\ell(p)=\mathbb E|Q|^2-1 there.

The Hessian in angular gaps

Label the critical values in counterclockwise order as v1,v2,v3v_1,v_2,v_3. Choose local real arguments θj(τ)\theta_j(\tau) of (5), and let ω1=(θ2−θ1)/(2π)\omega_1=(\theta_2-\theta_1)/(2\pi) and ω2=(θ3−θ2)/(2π)\omega_2=(\theta_3-\theta_2)/(2\pi); the last gap is 1−ω1−ω21-\omega_1-\omega_2. Put

δj=vj+1−vj,ϵj=vj+12−vj2,j=1,2.\delta_j=v_{j+1}-v_j,\qquad \epsilon_j=v_{j+1}^2-v_j^2,\qquad j=1,2.

The real Jacobian at zero is

J=1π(Im⁡δ1Re⁡δ1Im⁡δ2Re⁡δ2).(12)J=\frac1\pi \begin{pmatrix} \operatorname{Im}\delta_1&\operatorname{Re}\delta_1\\ \operatorname{Im}\delta_2&\operatorname{Re}\delta_2 \end{pmatrix}. \tag{12}

It is invertible: three distinct points of a circle are not collinear. Hence the gaps are real local coordinates for this family. Let gg be the two-component gradient of log⁡ℓ\log\ell in these coordinates at pp. By (11), it is determined by

JTg=(Re⁡m,−Im⁡m)T.(13)J^{\mathsf T}g=(\operatorname{Re}m,-\operatorname{Im}m)^{\mathsf T}. \tag{13}

Expanding the arguments in (5) gives

θj(τ)=θj(0)+2Im⁡(τvj)+Im⁡(τ2vj2)+O(∣τ∣3).\theta_j(\tau)=\theta_j(0)+2\operatorname{Im}(\tau v_j) +\operatorname{Im}(\tau^2v_j^2)+O(|\tau|^3).

Thus each ωj\omega_j is harmonic in τ\tau and its real Hessian at zero is

Hj=1π(Im⁡ϵjRe⁡ϵjRe⁡ϵj−Im⁡ϵj).(14)H_j=\frac1\pi \begin{pmatrix} \operatorname{Im}\epsilon_j&\operatorname{Re}\epsilon_j\\ \operatorname{Re}\epsilon_j&-\operatorname{Im}\epsilon_j \end{pmatrix}. \tag{14}

The chain rule says that the Hessian in the gaps is J−T(Dτ2u−g1H1−g2H2)J−1J^{-\mathsf T}(D^2_\tau u-g_1H_1-g_2H_2)J^{-1}. The middle symmetric matrix has trace Var⁡(Q)−1\operatorname{Var}(Q)-1 and eigenvalues

Var⁡(Q)−12 ± ∣B∣,B=EW+EQ2−m22+iπ∑j=12gjϵj.(15)\frac{\operatorname{Var}(Q)-1}{2}\ \mathbin{\pm}\ |B|, \quad B=\mathbb E W+\frac{\mathbb E Q^2-m^2}{2} +\frac{i}{\pi}\sum_{j=1}^2g_j\epsilon_j. \tag{15}

Congruence by an invertible real matrix preserves positive semidefiniteness. Therefore convexity of log⁡ℓ\log\ell in the gaps at pp is equivalent to the exact inequality

Var⁡(Q)−1≥2∣B∣.(16)\boxed{\operatorname{Var}(Q)-1\ge2|B|.} \tag{16}

Equation (16) is a criterion, not an established inequality for every quartic. The weaker trace inequality and the stationary-point test following (11) are likewise unresolved in the linked research sketch.