Erdős Problem #973

Erdős Problem 973 --- collection record

PROVENANCE AND STATUS

Saved collaborative-database status: open. Snapshot checked: 2026-09-23T17:22:24Z. Latest status and formalization links: https://teorth.github.io/erdosproblems/

This is an attributed collection of existing writeups. The model-folder names below identify their original attribution; placement in GPT_6_Astra_Ultra does not attribute their mathematical work to that model. All locally available source versions selected for this problem are reproduced below, without editing their text.

Proof claims, computations, literature statements, and completion estimates in the reused text have not been independently verified in this collection pass. A source claiming a solution does not change the saved upstream status.

Primary source (latest numbered version in the preferred model folder): attacks/erdos/gpt_pro_5.2/973.tex.

REUSED SOURCE 1

Does there exist a constant $C>1$ such that, for every $n\geq 2$, there exists a sequence $z_i\in \mathbb{C}$ with $z_1=1$ and $\lvert z_i\rvert \geq 1$ for all $1\leq i\leq n$ with\[\max_{2\leq k\leq n+1}\left\lvert \sum_{1\leq i\leq n}z_i^k\right\rvert < C^{-n}?\] This is Problem 7.3 in [Ha74], where it is attributed to Erdős. Erdős proved (as described on p.35 of [Tu84b]) that such a sequence does exist with $\lvert z_i\rvert\leq 1$. Indeed, Erdős' construction gives a value of $C\approx 1.32$. In [Er92f] (a different) Erdős refines this analysis, proving that if\[M_2=\min_{z_j} \max_{2\leq k\leq n+1} \left\lvert \sum_{1\leq j\leq n}z_j^k\right\rvert,\]where the minimum is take over all $z_j\in \mathbb{C}$ with $\max \lvert z_j\rvert=1$, then\[(1.746)^{-n} < M_2 < (1.745)^{-n}.\]Tang notes in the comments that Theorem 6.1 of [Tu84b] implies that, if $\lvert z_i\rvert \geq 1$ for all $i$, then\[\max_{2\leq k\leq n+1}\left\lvert \sum_{1\leq i\leq n}z_i^k\right\rvert \geq (2e)^{-(1+o(1))n}.\]See also [519] . References [Er92f] Erdős, L., On some problems of {P}. {T}urán concerning power sums of complex numbers . Acta Math. Hungar. $1992$, 11--24. [Ha74] Hayman, W. K., Research problems in function theory: new problems . $1974$, 155--180. [Tu84b] Turán, Paul, On a new method of analysis and its applications . $1984$, xvi+584.

1) FORMAL RESTATEMENT
Question: Does there exist an absolute constant $C>1$ such that for every integer $n\ge 2$ there exist complex numbers $z_1,\dots,z_n$ with \[ z_1=1,\qquad |z_i|\ge 1\ (1\le i\le n), \] and such that \[ \max_{2\le k\le n+1}\Big|\sum_{i=1}^n z_i^k\Big| < C^{-n}? \]

2) QUICK LITERATURE/CONTEXT CHECK
The statement records that Erdős produced sequences under the opposite constraint $|z_i|\le 1$ (and more generally $\max|z_i|=1$) with exponentially small power sums, while Turán's theorem gives an exponential lower bound when $|z_i|\ge 1$. I do not use any other external input.

3) ATTACK PLAN
Proof track: attempt an explicit construction meeting $|z_i|\ge 1$ that forces the first $n$ power sums to be exponentially small; investigate whether one can restrict attention to $|z_i|=1$ without loss, or whether varying radii helps.

Disproof track: attempt to prove a universal lower bound of the form $\max_{2\le k\le n+1}|\sum z_i^k|\ge c^{-n}$ for some $c<1$ that is too large to allow any fixed $C>1$; this seems incompatible with the stated Turán lower bound (which is also exponential), so a disproof would require sharpening.

I only manage a complete analysis for the smallest case $n=2$.

4) WORK
FAST REALITY CHECK (the case $n=2$). For $n=2$ and $z_1=1$, writing $z_2=re^{i\theta}$, I performed a coarse grid search over $r\in[1,4]$ and $\theta$ on a $2000$-point mesh. The minimal value found for \[ \max\big(|1+z_2^2|,\ |1+z_2^3|\big) \] was $0.6180339887\dots$ attained at $r=1$ and $\theta=2\pi/5$. The next proposition proves this is the true optimum.

Lemma 973.1 (monotonicity in the radius for a single term). Fix an integer $k\ge 1$ and an angle $\theta\in\mathbb{R}$. For $r\ge 1$, define \[ F_{k,\theta}(r):=\big|1+r^k e^{ik\theta}\big|. \] Then $F_{k,\theta}(r)$ is nondecreasing for $r\ge 1$.

Proof. Compute the squared modulus: \[ F_{k,\theta}(r)^2 = \big(1+r^k e^{ik\theta}\big)\big(1+r^k e^{-ik\theta}\big)=1+r^{2k}+2r^k\cos(k\theta). \] Differentiate for $r>0$: \[ \frac{d}{dr}F_{k,\theta}(r)^2 = 2k r^{2k-1} + 2k r^{k-1}\cos(k\theta) =2k r^{k-1}\big(r^k+\cos(k\theta)\big). \] For $r\ge 1$, we have $r^k\ge 1$ and $\cos(k\theta)\ge -1$, hence $r^k+\cos(k\theta)\ge 0$, so the derivative is nonnegative. Thus $F_{k,\theta}(r)^2$ is nondecreasing for $r\ge 1$, and therefore so is $F_{k,\theta}(r)=\sqrt{F_{k,\theta}(r)^2}$. $\square$

Proposition 973.2 (exact minimax for $n=2$). Let $n=2$, with $z_1=1$ and $z_2\in\mathbb{C}$ satisfying $|z_2|\ge 1$. Define \[ M_2^{(\ge 1)}:=\inf_{|z_2|\ge 1}\ \max\big(|1+z_2^2|,\ |1+z_2^3|\big). \] Then \[ M_2^{(\ge 1)}=\frac{\sqrt5-1}{2}, \] and the infimum is attained, for instance, at $z_2=e^{2\pi i/5}$ (and by symmetry at $e^{-2\pi i/5}$).

Proof. Write $z_2=re^{i\theta}$ with $r\ge 1$. By Lemma lem:973-radius, for each fixed $\theta$ the quantities $|1+z_2^2|=|1+r^2 e^{2i\theta}|$ and $|1+z_2^3|=|1+r^3 e^{3i\theta}|$ are both nondecreasing in $r$ for $r\ge 1$. Therefore their maximum is minimized at the smallest allowed radius $r=1$. Hence \[ M_2^{(\ge 1)} = \inf_{\theta\in\mathbb{R}} \max\big(|1+e^{2i\theta}|,\ |1+e^{3i\theta}|\big). \] Using $|1+e^{i\varphi}|=2|\cos(\varphi/2)|$, we rewrite \[ |1+e^{2i\theta}|=2|\cos\theta|,\qquad |1+e^{3i\theta}|=2|\cos(3\theta/2)|. \] Let \[ m(\theta):=\max\big(|\cos\theta|,\ |\cos(3\theta/2)|\big),\qquad\text{so }M_2^{(\ge 1)}=2\inf_{\theta} m(\theta). \] The function $m$ is $2\pi$-periodic and even, so it suffices to minimize over $\theta\in[0,\pi]$. At any minimizer $\theta_*$, one must have $|\cos\theta_*|=|\cos(3\theta_*/2)|$; otherwise if (say) $|\cos\theta_*|>|\cos(3\theta_*/2)|$ then by continuity one can perturb $\theta_*$ slightly to decrease $|\cos\theta|$ while keeping $|\cos(3\theta/2)|$ strictly below the previous maximum, reducing $m$, contradicting minimality. Thus we solve the equation \[ |\cos\theta|=|\cos(3\theta/2)|. \] There are two cases.

Case 1: $\cos\theta=\cos(3\theta/2)$. Then either $\theta\equiv 3\theta/2\pmod{2\pi}$ (giving $\theta\equiv 0$) or $\theta\equiv -3\theta/2\pmod{2\pi}$ (giving $5\theta/2\equiv 0\pmod{2\pi}$, hence $\theta\in\{0,4\pi/5\}$ within $[0,\pi]$). At $\theta=0$, $m(0)=1$. At $\theta=4\pi/5$, $|\cos\theta|=|\cos(4\pi/5)|=|\cos(\pi-\pi/5)|=\cos(\pi/5)=\frac{\sqrt5+1}{4}\approx 0.809$. So this case yields $2m\ge 2\cdot \frac{\sqrt5+1}{4}=\frac{\sqrt5+1}{2}\approx 1.618$.

Case 2: $\cos\theta=-\cos(3\theta/2)$. Using $\cos a+\cos b=2\cos\frac{a+b}{2}\cos\frac{a-b}{2}$, the equation becomes \[ \cos\theta+\cos(3\theta/2)=0 \iff 2\cos\Big(\frac{5\theta}{4}\Big)\cos\Big(\frac{\theta}{4}\Big)=0. \] Within $\theta\in[0,\pi]$, the factor $\cos(\theta/4)$ is positive, so we must have $\cos(5\theta/4)=0$, i.e. \[ \frac{5\theta}{4}=\frac{\pi}{2}+\pi m \quad\Longrightarrow\quad \theta=\frac{2\pi}{5}(1+2m). \] The unique solution in $[0,\pi]$ is $\theta=2\pi/5$. At this angle, \[ |\cos\theta|=\cos(2\pi/5)=\cos(72^\circ)=\frac{\sqrt5-1}{4}. \] Hence $m(2\pi/5)=\frac{\sqrt5-1}{4}$ and the corresponding objective value is \[ 2m(2\pi/5)=\frac{\sqrt5-1}{2}. \] Comparing with Case 1 (and boundary values), this is the global minimum. Thus $M_2^{(\ge 1)}=(\sqrt5-1)/2$, achieved at $z_2=e^{i\theta}=e^{2\pi i/5}$. $\square$

5) VERIFICATION
Lemma lem:973-radius was checked by explicit differentiation; the key inequality $r^k+\cos(k\theta)\ge r^k-1\ge 0$ holds for $r\ge 1$. In Proposition prop:973-n2, the reduction to $r=1$ is valid because both competing terms are separately nondecreasing in $r$ for fixed $\theta$. The trigonometric minimization checks all solutions to the equal-max condition and compares values.

6) FINAL
UNRESOLVED

(i) Strongest proved partial result. For $n=2$ the exact minimax value under the constraint $|z_2|\ge 1$ is $M_2^{(\ge 1)}=(\sqrt5-1)/2$ (Proposition prop:973-n2), achieved on the unit circle. Also, for each individual term $|1+r^k e^{ik\theta}|$ is minimized at the smallest allowed radius $r=1$ (Lemma lem:973-radius).

(ii) First gap (crisp). For general $n$, construct (or rule out) sequences $z_1=1$, $|z_i|\ge 1$ such that \[ \max_{2\le k\le n+1}\Big|\sum_{i=1}^n z_i^k\Big|\le C^{-n} \] for some absolute constant $C>1$ independent of $n$.

(iii) Top 3 next moves.

• Determine whether one may assume $|z_i|=1$ without loss for the minimax problem under $|z_i|\ge 1$; Lemma lem:973-radius shows this for $n=2$ but not for coupled sums.
• Attempt an explicit construction: choose $z_i$ as (possibly repeated) points on or outside the unit circle designed to make the first $n$ power sums small simultaneously (e.g. via polynomial/Newton-identity control of power sums), while keeping $|z_i|\ge 1$.
• Compute numerical optima for small $n$ with constrained optimization (e.g. restricting to $|z_i|=1$ and varying phases) to guess the correct exponential base for the $|z_i|\ge 1$ constraint.

(iv) Minimal counterexample structure. A negative answer would mean: for every $C>1$ there are infinitely many $n$ such that for all choices with $z_1=1$ and $|z_i|\ge 1$, \[ \max_{2\le k\le n+1}\Big|\sum_{i=1}^n z_i^k\Big|\ge C^{-n}. \] In other words the minimax quantity for the $|z_i|\ge 1$ constraint would decay at best subexponentially (or with base $\le 1$), contradicting the heuristic suggested by the stated Turán-type exponential lower bound; any such counterexample would likely hinge on a coupled constraint preventing simultaneous cancellation across the $n$ different exponents $k=2,\dots,n+1$.

COMPLETION ESTIMATE

Newly verified progress in this collection pass: $0\%$. This measures new verification work, not the amount of mathematics achieved in the reused text.

Does there exist a constant $C>1$ such that, for every $n\geq 2$, there exists a sequence $z_i\in \mathbb{C}$ with $z_1=1$ and $\lvert z_i\rvert \geq 1$ for all $1\leq i\leq n$ with\[\max_{2\leq k\leq n+1}\left\lvert \sum_{1\leq i\leq n}z_i^k\right\rvert < C^{-n}?\] This is Problem 7.3 in [Ha74], where it is attributed to Erdős. Erdős proved (as described on p.35 of [Tu84b]) that such a sequence does exist with $\lvert z_i\rvert\leq 1$. Indeed, Erdős' construction gives a value of $C\approx 1.32$. In [Er92f] (a different) Erdős refines this analysis, proving that if\[M_2=\min_{z_j} \max_{2\leq k\leq n+1} \left\lvert \sum_{1\leq j\leq n}z_j^k\right\rvert,\]where the minimum is take over all $z_j\in \mathbb{C}$ with $\max \lvert z_j\rvert=1$, then\[(1.746)^{-n} < M_2 < (1.745)^{-n}.\]Tang notes in the comments that Theorem 6.1 of [Tu84b] implies that, if $\lvert z_i\rvert \geq 1$ for all $i$, then\[\max_{2\leq k\leq n+1}\left\lvert \sum_{1\leq i\leq n}z_i^k\right\rvert \geq (2e)^{-(1+o(1))n}.\]See also [519] . References [Er92f] Erdős, L., On some problems of {P}. {T}urán concerning power sums of complex numbers . Acta Math. Hungar. $1992$, 11--24. [Ha74] Hayman, W. K., Research problems in function theory: new problems . $1974$, 155--180. [Tu84b] Turán, Paul, On a new method of analysis and its applications . $1984$, xvi+584.

1) FORMAL RESTATEMENT
Question: Does there exist an absolute constant $C>1$ such that for every integer $n\ge 2$ there exist complex numbers $z_1,\dots,z_n$ with \[ z_1=1,\qquad |z_i|\ge 1\ (1\le i\le n), \] and such that \[ \max_{2\le k\le n+1}\Big|\sum_{i=1}^n z_i^k\Big| < C^{-n}? \]

2) QUICK LITERATURE/CONTEXT CHECK
The statement records that Erdős produced sequences under the opposite constraint $|z_i|\le 1$ (and more generally $\max|z_i|=1$) with exponentially small power sums, while Turán's theorem gives an exponential lower bound when $|z_i|\ge 1$. I do not use any other external input.

3) ATTACK PLAN
Proof track: attempt an explicit construction meeting $|z_i|\ge 1$ that forces the first $n$ power sums to be exponentially small; investigate whether one can restrict attention to $|z_i|=1$ without loss, or whether varying radii helps.

Disproof track: attempt to prove a universal lower bound of the form $\max_{2\le k\le n+1}|\sum z_i^k|\ge c^{-n}$ for some $c<1$ that is too large to allow any fixed $C>1$; this seems incompatible with the stated Turán lower bound (which is also exponential), so a disproof would require sharpening.

I only manage a complete analysis for the smallest case $n=2$.

4) WORK
FAST REALITY CHECK (the case $n=2$). For $n=2$ and $z_1=1$, writing $z_2=re^{i\theta}$, I performed a coarse grid search over $r\in[1,4]$ and $\theta$ on a $2000$-point mesh. The minimal value found for \[ \max\big(|1+z_2^2|,\ |1+z_2^3|\big) \] was $0.6180339887\dots$ attained at $r=1$ and $\theta=2\pi/5$. The next proposition proves this is the true optimum.

Lemma 973.1 (monotonicity in the radius for a single term). Fix an integer $k\ge 1$ and an angle $\theta\in\mathbb{R}$. For $r\ge 1$, define \[ F_{k,\theta}(r):=\big|1+r^k e^{ik\theta}\big|. \] Then $F_{k,\theta}(r)$ is nondecreasing for $r\ge 1$.

Proof. Compute the squared modulus: \[ F_{k,\theta}(r)^2 = \big(1+r^k e^{ik\theta}\big)\big(1+r^k e^{-ik\theta}\big)=1+r^{2k}+2r^k\cos(k\theta). \] Differentiate for $r>0$: \[ \frac{d}{dr}F_{k,\theta}(r)^2 = 2k r^{2k-1} + 2k r^{k-1}\cos(k\theta) =2k r^{k-1}\big(r^k+\cos(k\theta)\big). \] For $r\ge 1$, we have $r^k\ge 1$ and $\cos(k\theta)\ge -1$, hence $r^k+\cos(k\theta)\ge 0$, so the derivative is nonnegative. Thus $F_{k,\theta}(r)^2$ is nondecreasing for $r\ge 1$, and therefore so is $F_{k,\theta}(r)=\sqrt{F_{k,\theta}(r)^2}$. $\square$

Proposition 973.2 (exact minimax for $n=2$). Let $n=2$, with $z_1=1$ and $z_2\in\mathbb{C}$ satisfying $|z_2|\ge 1$. Define \[ M_2^{(\ge 1)}:=\inf_{|z_2|\ge 1}\ \max\big(|1+z_2^2|,\ |1+z_2^3|\big). \] Then \[ M_2^{(\ge 1)}=\frac{\sqrt5-1}{2}, \] and the infimum is attained, for instance, at $z_2=e^{2\pi i/5}$ (and by symmetry at $e^{-2\pi i/5}$).

Proof. Write $z_2=re^{i\theta}$ with $r\ge 1$. By Lemma lem:973-radius, for each fixed $\theta$ the quantities $|1+z_2^2|=|1+r^2 e^{2i\theta}|$ and $|1+z_2^3|=|1+r^3 e^{3i\theta}|$ are both nondecreasing in $r$ for $r\ge 1$. Therefore their maximum is minimized at the smallest allowed radius $r=1$. Hence \[ M_2^{(\ge 1)} = \inf_{\theta\in\mathbb{R}} \max\big(|1+e^{2i\theta}|,\ |1+e^{3i\theta}|\big). \] Using $|1+e^{i\varphi}|=2|\cos(\varphi/2)|$, we rewrite \[ |1+e^{2i\theta}|=2|\cos\theta|,\qquad |1+e^{3i\theta}|=2|\cos(3\theta/2)|. \] Let \[ m(\theta):=\max\big(|\cos\theta|,\ |\cos(3\theta/2)|\big),\qquad\text{so }M_2^{(\ge 1)}=2\inf_{\theta} m(\theta). \] The function $m$ is $2\pi$-periodic and even, so it suffices to minimize over $\theta\in[0,\pi]$. At any minimizer $\theta_*$, one must have $|\cos\theta_*|=|\cos(3\theta_*/2)|$; otherwise if (say) $|\cos\theta_*|>|\cos(3\theta_*/2)|$ then by continuity one can perturb $\theta_*$ slightly to decrease $|\cos\theta|$ while keeping $|\cos(3\theta/2)|$ strictly below the previous maximum, reducing $m$, contradicting minimality. Thus we solve the equation \[ |\cos\theta|=|\cos(3\theta/2)|. \] There are two cases.

Case 1: $\cos\theta=\cos(3\theta/2)$. Then either $\theta\equiv 3\theta/2\pmod{2\pi}$ (giving $\theta\equiv 0$) or $\theta\equiv -3\theta/2\pmod{2\pi}$ (giving $5\theta/2\equiv 0\pmod{2\pi}$, hence $\theta\in\{0,4\pi/5\}$ within $[0,\pi]$). At $\theta=0$, $m(0)=1$. At $\theta=4\pi/5$, $|\cos\theta|=|\cos(4\pi/5)|=|\cos(\pi-\pi/5)|=\cos(\pi/5)=\frac{\sqrt5+1}{4}\approx 0.809$. So this case yields $2m\ge 2\cdot \frac{\sqrt5+1}{4}=\frac{\sqrt5+1}{2}\approx 1.618$.

Case 2: $\cos\theta=-\cos(3\theta/2)$. Using $\cos a+\cos b=2\cos\frac{a+b}{2}\cos\frac{a-b}{2}$, the equation becomes \[ \cos\theta+\cos(3\theta/2)=0 \iff 2\cos\Big(\frac{5\theta}{4}\Big)\cos\Big(\frac{\theta}{4}\Big)=0. \] Within $\theta\in[0,\pi]$, the factor $\cos(\theta/4)$ is positive, so we must have $\cos(5\theta/4)=0$, i.e. \[ \frac{5\theta}{4}=\frac{\pi}{2}+\pi m \quad\Longrightarrow\quad \theta=\frac{2\pi}{5}(1+2m). \] The unique solution in $[0,\pi]$ is $\theta=2\pi/5$. At this angle, \[ |\cos\theta|=\cos(2\pi/5)=\cos(72^\circ)=\frac{\sqrt5-1}{4}. \] Hence $m(2\pi/5)=\frac{\sqrt5-1}{4}$ and the corresponding objective value is \[ 2m(2\pi/5)=\frac{\sqrt5-1}{2}. \] Comparing with Case 1 (and boundary values), this is the global minimum. Thus $M_2^{(\ge 1)}=(\sqrt5-1)/2$, achieved at $z_2=e^{i\theta}=e^{2\pi i/5}$. $\square$

5) VERIFICATION
Lemma lem:973-radius was checked by explicit differentiation; the key inequality $r^k+\cos(k\theta)\ge r^k-1\ge 0$ holds for $r\ge 1$. In Proposition prop:973-n2, the reduction to $r=1$ is valid because both competing terms are separately nondecreasing in $r$ for fixed $\theta$. The trigonometric minimization checks all solutions to the equal-max condition and compares values.

6) FINAL
UNRESOLVED

(i) Strongest proved partial result. For $n=2$ the exact minimax value under the constraint $|z_2|\ge 1$ is $M_2^{(\ge 1)}=(\sqrt5-1)/2$ (Proposition prop:973-n2), achieved on the unit circle. Also, for each individual term $|1+r^k e^{ik\theta}|$ is minimized at the smallest allowed radius $r=1$ (Lemma lem:973-radius).

(ii) First gap (crisp). For general $n$, construct (or rule out) sequences $z_1=1$, $|z_i|\ge 1$ such that \[ \max_{2\le k\le n+1}\Big|\sum_{i=1}^n z_i^k\Big|\le C^{-n} \] for some absolute constant $C>1$ independent of $n$.

(iii) Top 3 next moves.

• Determine whether one may assume $|z_i|=1$ without loss for the minimax problem under $|z_i|\ge 1$; Lemma lem:973-radius shows this for $n=2$ but not for coupled sums.
• Attempt an explicit construction: choose $z_i$ as (possibly repeated) points on or outside the unit circle designed to make the first $n$ power sums small simultaneously (e.g. via polynomial/Newton-identity control of power sums), while keeping $|z_i|\ge 1$.
• Compute numerical optima for small $n$ with constrained optimization (e.g. restricting to $|z_i|=1$ and varying phases) to guess the correct exponential base for the $|z_i|\ge 1$ constraint.

(iv) Minimal counterexample structure. A negative answer would mean: for every $C>1$ there are infinitely many $n$ such that for all choices with $z_1=1$ and $|z_i|\ge 1$, \[ \max_{2\le k\le n+1}\Big|\sum_{i=1}^n z_i^k\Big|\ge C^{-n}. \] In other words the minimax quantity for the $|z_i|\ge 1$ constraint would decay at best subexponentially (or with base $\le 1$), contradicting the heuristic suggested by the stated Turán-type exponential lower bound; any such counterexample would likely hinge on a coupled constraint preventing simultaneous cancellation across the $n$ different exponents $k=2,\dots,n+1$.