Erdős Problem #671

Erdős Problem 671 --- 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_5.2/671.tex.

REUSED SOURCE 1

1) FORMAL RESTATEMENT

Fix, for each integer $n\ge 1$, a list of pairwise distinct nodes \[ a_1^n,\dots,a_n^n\in[-1,1]. \] For each $n$ and each $1\le i\le n$, let $p_i^n\in\mathbb R[x]$ denote the unique polynomial of degree at most $n-1$ such that \[ p_i^n(a_i^n)=1,\qquad p_i^n(a_{i'}^n)=0\ \ (1\le i'\le n,\ i'\ne i). \] For each function $f:[-1,1]\to\mathbb R$ define the (degree $\le n-1$) Lagrange interpolation polynomial \[ (\mathcal L^n f)(x)=\sum_{i=1}^n f(a_i^n)\,p_i^n(x). \] Define the Lebesgue function \[ \Lambda_n(x):=\sum_{i=1}^n |p_i^n(x)|\qquad(x\in[-1,1]). \] Question $Q1$: Does there exist a choice of nodes $\{a_i^n\}$ such that for every continuous $f\in C([-1,1])$ there exists a point $x\in[-1,1]$ (depending on $f$) with \[ \limsup_{n\to\infty} \Lambda_n(x)=\infty\quad\text{and}\quad (\mathcal L^n f)(x)\to f(x)\ \text{as }n\to\infty? \] Question $Q2$: Does there exist a choice of nodes $\{a_i^n\}$ such that \[ \limsup_{n\to\infty}\Lambda_n(x)=\infty\quad\text{for every }x\in[-1,1], \] and yet for every continuous $f\in C([-1,1])$ there exists some $x\in[-1,1]$ (depending on $f$) with $(\mathcal L^n f)(x)\to f(x)$?

Edge cases/conventions: $\mathcal L^n f$ is well-defined for any $f$ by only using the values $f(a_i^n)$. For $n=1$ we have $p_1^1\equiv 1$ and $\Lambda_1\equiv 1$.

2) QUICK LITERATURE/CONTEXT CHECK

As stated in the problem text: - Bernstein $1931$ proved that for any choice of nodes there exists $x_0\in[-1,1]$ with $\limsup_{n\to\infty}\Lambda_n(x_0)=\infty$. - Erdős--Vértesi $1980$ proved that for any choice of nodes there exists a continuous $f$ such that $\limsup_{n\to\infty}|(\mathcal L^n f)(x)|=\infty$ for almost all $x\in[-1,1]$.

I do not use any other external results here.

3) ATTACK PLAN

Proof-oriented ideas: - Relate $\Lambda_n(x)$ to the operator norm of the evaluation functional $f\mapsto (\mathcal L^n f)(x)$ and use functional-analytic tools (uniform boundedness) to produce divergence examples at fixed $x$. - Use polynomial approximation: since $\mathcal L^n$ reproduces polynomials of degree $\le n-1$, the interpolation error at $x$ is controlled by $(1+\Lambda_n(x))$ times the best uniform approximation error $E_{n-1}(f)$.

Disproof-oriented ideas: - Try to show that for any node choice and for any set $S\subset[-1,1]$ on which $\limsup\Lambda_n(x)=\infty$, there exists continuous $f$ such that $\mathcal L^n f(x)$ fails to converge for all $x\in S$ (this would negate $Q1$ or $Q2$ depending on $S$).

I do not complete either track.

4) WORK

FAST REALITY CHECK (small $n$, explicit computations)

For two standard node choices, I computed $\Lambda_n(x)$ via barycentric Lagrange evaluation.

- Equispaced nodes: $a_i^n=-1+2(i-1)/(n-1)$. - Chebyshev nodes (first kind): $a_i^n=\cos\left(\pi(2i-1)/(2n)\right)$.

Selected numerical values (approximate): \[ \begin{array}{c|cc|cc} n & \Lambda_n(0)\ \text{(equispaced)} & \sup_{x\in[-1,1]}\Lambda_n(x)\ \text{(grid)} & \Lambda_n(0)\ \text{(Chebyshev)} & \sup_x\Lambda_n(x)\ \text{(grid)}\\\hline 4 & 1.25 & 1.63 & 1.414 & 1.848\\ 8 & 1.488 & 6.93 & 1.848 & 2.287\\ 12 & 1.624 & 51.21 & 2.104 & 2.545\\ 16 & 1.718 & 512.35 & 2.287 & 2.728 \end{array} \] For Chebyshev nodes at $x=0$, restricting to even $n$ gives slow growth: $\Lambda_{10}(0)\approx1.99$, $\Lambda_{20}(0)\approx2.43$, $\Lambda_{40}(0)\approx2.87$, $\Lambda_{80}(0)\approx3.31$, $\Lambda_{160}(0)\approx3.75$.

These sanity checks confirm that $\Lambda_n(x)$ can diverge very fast (equispaced) or slowly (Chebyshev) while still having $\limsup\Lambda_n(x)=\infty$.

Lemma 671.1 (partition of unity and reproduction)

For each $n$ we have \[ \sum_{i=1}^n p_i^n(x)\equiv 1\quad\text{as polynomials in }x. \] Moreover, for every polynomial $q\in\mathbb R[x]$ with $\deg q\le n-1$ we have \[ \mathcal L^n q\equiv q. \]

Proof. Define $r(x):=\sum_{i=1}^n p_i^n(x)-1$. Then $\deg r\le n-1$ and for each node $a_j^n$ we have \[ r(a_j^n)=\sum_{i=1}^n p_i^n(a_j^n)-1 = 1-1=0, \] because $p_j^n(a_j^n)=1$ and $p_i^n(a_j^n)=0$ for $i\ne j$. Thus $r$ has $n$ distinct zeros and degree at most $n-1$, forcing $r\equiv 0$.

For the reproduction statement, fix $q$ with $\deg q\le n-1$ and define \[ s(x):=q(x)-\sum_{i=1}^n q(a_i^n)p_i^n(x)=q(x)-(\mathcal L^n q)(x). \] Then $\deg s\le n-1$. For each $j$, \[ s(a_j^n)=q(a_j^n)-\sum_{i=1}^n q(a_i^n)p_i^n(a_j^n)=q(a_j^n)-q(a_j^n)=0. \] So $s$ has $n$ distinct zeros and degree at most $n-1$, hence $s\equiv 0$ and $\mathcal L^n q\equiv q$.

Lemma 671.2 (operator norm identity)

Fix $n\ge 1$ and $x\in[-1,1]$. Define the bounded linear functional \[ T_{n,x}:C([-1,1])\to\mathbb R,\qquad T_{n,x}(f):=(\mathcal L^n f)(x). \] Then \[ |T_{n,x}(f)|\le \Lambda_n(x)\,\|f\|_\infty\quad\text{for all }f\in C([-1,1]), \] and the operator norm satisfies \[ \|T_{n,x}\|=\sup_{\|f\|_\infty\le 1}|T_{n,x}(f)|=\Lambda_n(x). \]

Proof. From the definition, \[ T_{n,x}(f)=\sum_{i=1}^n f(a_i^n)p_i^n(x). \] Taking absolute values and using $|f(a_i^n)|\le\|f\|_\infty$ gives \[ |T_{n,x}(f)|\le \sum_{i=1}^n |f(a_i^n)|\,|p_i^n(x)|\le \|f\|_\infty\sum_{i=1}^n |p_i^n(x)|=\Lambda_n(x)\|f\|_\infty. \] So $\|T_{n,x}\|\le\Lambda_n(x)$.

For the reverse inequality, define a function $g$ on the finite node set by \[ g(a_i^n):=\begin{cases} 1,& p_i^n(x)\ge 0,\\ -1,& p_i^n(x)<0. \end{cases} \] Then $|g(a_i^n)|=1$ for all $i$. Since the node set is finite, we can extend these prescribed values to a continuous function $f\in C([-1,1])$ with $\|f\|_\infty\le 1$ (for instance: sort the nodes in increasing order and define $f$ piecewise linearly so that $f$ takes the value $g(a_i^n)$ at each node; this produces a continuous $f$ with values in $[-1,1]$).

For this $f$, we have \[ T_{n,x}(f)=\sum_{i=1}^n f(a_i^n)p_i^n(x)=\sum_{i=1}^n \operatorname{sgn}(p_i^n(x))\,p_i^n(x)=\sum_{i=1}^n |p_i^n(x)|=\Lambda_n(x). \] Hence $\|T_{n,x}\|\ge \Lambda_n(x)$. Combining with the upper bound yields $\|T_{n,x}\|=\Lambda_n(x)$.

Lemma 671.3 (error bound via best polynomial approximation)

For $f\in C([-1,1])$ define the best uniform approximation error by degree $\le n-1$ polynomials: \[ E_{n-1}(f):=\inf\{\|f-q\|_\infty: q\in\mathbb R[x],\ \deg q\le n-1\}. \] Then for every $x\in[-1,1]$, \[ |(\mathcal L^n f)(x)-f(x)|\le (1+\Lambda_n(x))\,E_{n-1}(f). \]

Proof. Fix any polynomial $q$ with $\deg q\le n-1$. By Lemma 671.1, $\mathcal L^n q=q$. Thus \[ (\mathcal L^n f)(x)-f(x)=(\mathcal L^n (f-q))(x)-(f-q)(x). \] Taking absolute values and applying the triangle inequality, \[ |(\mathcal L^n f)(x)-f(x)|\le |(\mathcal L^n (f-q))(x)|+|(f-q)(x)|. \] By Lemma 671.2 (the bound $|T_{n,x}(h)|\le\Lambda_n(x)\|h\|_\infty$ applied to $h=f-q$), \[ |(\mathcal L^n (f-q))(x)|\le\Lambda_n(x)\,\|f-q\|_\infty. \] Also $|(f-q)(x)|\le\|f-q\|_\infty$. Therefore \[ |(\mathcal L^n f)(x)-f(x)|\le (\Lambda_n(x)+1)\,\|f-q\|_\infty. \] Taking the infimum over all such $q$ gives the claimed inequality.

Proposition 671.4 (uniform boundedness at a fixed point)

Fix a point $x\in[-1,1]$. If $\sup_n \Lambda_n(x)=\infty$, then there exists $f\in C([-1,1])$ such that \[ \sup_n |(\mathcal L^n f)(x)|=\infty. \] In particular, $(\mathcal L^n f)(x)$ does not converge.

Proof. By Lemma 671.2, $\|T_{n,x}\|=\Lambda_n(x)$. Thus $\sup_n\|T_{n,x}\|=\infty$. The Banach space $C([-1,1])$ is complete under $\|\cdot\|_\infty$, and $\{T_{n,x}\}$ is a family of bounded linear functionals on it. By the Uniform Boundedness Principle (Banach--Steinhaus theorem), if $\sup_n\|T_{n,x}\|=\infty$ then there exists $f\in C([-1,1])$ with $\sup_n |T_{n,x}(f)|=\infty$. This is exactly the displayed statement.

5) VERIFICATION

- Lemma 671.1: checked that the argument uses only (i) degree bound $\le n-1$, (ii) $n$ distinct nodes. No hidden assumptions. - Lemma 671.2: the only nontrivial step is extending prescribed values on finitely many points to a continuous function with sup-norm $\le 1$. The explicit piecewise-linear construction works because $[-1,1]$ is an interval and the node set is finite. - Lemma 671.3: uses only Lemma 671.1 and the bound from Lemma 671.2. - Proposition 671.4: assumes the Uniform Boundedness Principle; this is a standard theorem of functional analysis. If one wishes to avoid it, one could reproduce its proof, but I did not.

Boundary cases: for $n=1$, $\Lambda_1(x)=1$ and all inequalities are trivially valid.

6) FINAL

UNRESOLVED

(i) Strongest proved partial result:

For any node choice and any $f\in C([-1,1])$, \[ |(\mathcal L^n f)(x)-f(x)|\le (1+\Lambda_n(x))E_{n-1}(f)\quad\text{for every }x\in[-1,1]. \] Also, for each fixed $x$, if $\sup_n\Lambda_n(x)=\infty$ then there exists $f\in C([-1,1])$ such that $(\mathcal L^n f)(x)$ is unbounded (hence diverges) at that same $x$.

(ii) First gap (crisp):

Given an arbitrary continuous $f$, must there exist a point $x$ with $\limsup_n\Lambda_n(x)=\infty$ such that nevertheless $(\mathcal L^n f)(x)\to f(x)$? I do not have a method to force convergence at a point where $\Lambda_n(x)$ is unbounded.

(iii) Top 3 next moves:

1. Attempt a Baire-category argument on $C([-1,1])$ for fixed nodes, studying the set of $f$ for which $\mathcal L^n f(x)$ converges at a given $x$ versus the growth rate of $\Lambda_n(x)$. 2. For candidate nodes (e.g. Chebyshev-type), try to quantify pointwise growth of $\Lambda_n(x)$ and relate it to typical decay rates of $E_{n-1}(f)$ via explicit modulus of continuity estimates. 3. Computationally: for concrete node systems, numerically approximate the set of $x$ where $\Lambda_n(x)$ is "small" infinitely often, and test pointwise convergence for slowly-approximable continuous functions.

(iv) Minimal counterexample structure:

To refute $Q1$ for a given node system, one would want a continuous $f$ such that for every $x$ with $\limsup_n\Lambda_n(x)=\infty$, the sequence $(\mathcal L^n f)(x)$ fails to converge (ideally is unbounded). Proposition 671.4 produces, for each fixed $x$, some $f_x$ with divergence at that $x$, but I do not know how to construct a single $f$ that forces divergence simultaneously on the full (possibly large) set of divergence points.

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.

1) FORMAL RESTATEMENT

Fix, for each integer $n\ge 1$, a list of pairwise distinct nodes \[ a_1^n,\dots,a_n^n\in[-1,1]. \] For each $n$ and each $1\le i\le n$, let $p_i^n\in\mathbb R[x]$ denote the unique polynomial of degree at most $n-1$ such that \[ p_i^n(a_i^n)=1,\qquad p_i^n(a_{i'}^n)=0\ \ (1\le i'\le n,\ i'\ne i). \] For each function $f:[-1,1]\to\mathbb R$ define the (degree $\le n-1$) Lagrange interpolation polynomial \[ (\mathcal L^n f)(x)=\sum_{i=1}^n f(a_i^n)\,p_i^n(x). \] Define the Lebesgue function \[ \Lambda_n(x):=\sum_{i=1}^n |p_i^n(x)|\qquad(x\in[-1,1]). \] Question $Q1$: Does there exist a choice of nodes $\{a_i^n\}$ such that for every continuous $f\in C([-1,1])$ there exists a point $x\in[-1,1]$ (depending on $f$) with \[ \limsup_{n\to\infty} \Lambda_n(x)=\infty\quad\text{and}\quad (\mathcal L^n f)(x)\to f(x)\ \text{as }n\to\infty? \] Question $Q2$: Does there exist a choice of nodes $\{a_i^n\}$ such that \[ \limsup_{n\to\infty}\Lambda_n(x)=\infty\quad\text{for every }x\in[-1,1], \] and yet for every continuous $f\in C([-1,1])$ there exists some $x\in[-1,1]$ (depending on $f$) with $(\mathcal L^n f)(x)\to f(x)$?

Edge cases/conventions: $\mathcal L^n f$ is well-defined for any $f$ by only using the values $f(a_i^n)$. For $n=1$ we have $p_1^1\equiv 1$ and $\Lambda_1\equiv 1$.

2) QUICK LITERATURE/CONTEXT CHECK

As stated in the problem text: - Bernstein $1931$ proved that for any choice of nodes there exists $x_0\in[-1,1]$ with $\limsup_{n\to\infty}\Lambda_n(x_0)=\infty$. - Erdős--Vértesi $1980$ proved that for any choice of nodes there exists a continuous $f$ such that $\limsup_{n\to\infty}|(\mathcal L^n f)(x)|=\infty$ for almost all $x\in[-1,1]$.

I do not use any other external results here.

3) ATTACK PLAN

Proof-oriented ideas: - Relate $\Lambda_n(x)$ to the operator norm of the evaluation functional $f\mapsto (\mathcal L^n f)(x)$ and use functional-analytic tools (uniform boundedness) to produce divergence examples at fixed $x$. - Use polynomial approximation: since $\mathcal L^n$ reproduces polynomials of degree $\le n-1$, the interpolation error at $x$ is controlled by $(1+\Lambda_n(x))$ times the best uniform approximation error $E_{n-1}(f)$.

Disproof-oriented ideas: - Try to show that for any node choice and for any set $S\subset[-1,1]$ on which $\limsup\Lambda_n(x)=\infty$, there exists continuous $f$ such that $\mathcal L^n f(x)$ fails to converge for all $x\in S$ (this would negate $Q1$ or $Q2$ depending on $S$).

I do not complete either track.

4) WORK

FAST REALITY CHECK (small $n$, explicit computations)

For two standard node choices, I computed $\Lambda_n(x)$ via barycentric Lagrange evaluation.

- Equispaced nodes: $a_i^n=-1+2(i-1)/(n-1)$. - Chebyshev nodes (first kind): $a_i^n=\cos\left(\pi(2i-1)/(2n)\right)$.

Selected numerical values (approximate): \[ \begin{array}{c|cc|cc} n & \Lambda_n(0)\ \text{(equispaced)} & \sup_{x\in[-1,1]}\Lambda_n(x)\ \text{(grid)} & \Lambda_n(0)\ \text{(Chebyshev)} & \sup_x\Lambda_n(x)\ \text{(grid)}\\\hline 4 & 1.25 & 1.63 & 1.414 & 1.848\\ 8 & 1.488 & 6.93 & 1.848 & 2.287\\ 12 & 1.624 & 51.21 & 2.104 & 2.545\\ 16 & 1.718 & 512.35 & 2.287 & 2.728 \end{array} \] For Chebyshev nodes at $x=0$, restricting to even $n$ gives slow growth: $\Lambda_{10}(0)\approx1.99$, $\Lambda_{20}(0)\approx2.43$, $\Lambda_{40}(0)\approx2.87$, $\Lambda_{80}(0)\approx3.31$, $\Lambda_{160}(0)\approx3.75$.

These sanity checks confirm that $\Lambda_n(x)$ can diverge very fast (equispaced) or slowly (Chebyshev) while still having $\limsup\Lambda_n(x)=\infty$.

Lemma 671.1 (partition of unity and reproduction)

For each $n$ we have \[ \sum_{i=1}^n p_i^n(x)\equiv 1\quad\text{as polynomials in }x. \] Moreover, for every polynomial $q\in\mathbb R[x]$ with $\deg q\le n-1$ we have \[ \mathcal L^n q\equiv q. \]

Proof. Define $r(x):=\sum_{i=1}^n p_i^n(x)-1$. Then $\deg r\le n-1$ and for each node $a_j^n$ we have \[ r(a_j^n)=\sum_{i=1}^n p_i^n(a_j^n)-1 = 1-1=0, \] because $p_j^n(a_j^n)=1$ and $p_i^n(a_j^n)=0$ for $i\ne j$. Thus $r$ has $n$ distinct zeros and degree at most $n-1$, forcing $r\equiv 0$.

For the reproduction statement, fix $q$ with $\deg q\le n-1$ and define \[ s(x):=q(x)-\sum_{i=1}^n q(a_i^n)p_i^n(x)=q(x)-(\mathcal L^n q)(x). \] Then $\deg s\le n-1$. For each $j$, \[ s(a_j^n)=q(a_j^n)-\sum_{i=1}^n q(a_i^n)p_i^n(a_j^n)=q(a_j^n)-q(a_j^n)=0. \] So $s$ has $n$ distinct zeros and degree at most $n-1$, hence $s\equiv 0$ and $\mathcal L^n q\equiv q$.

Lemma 671.2 (operator norm identity)

Fix $n\ge 1$ and $x\in[-1,1]$. Define the bounded linear functional \[ T_{n,x}:C([-1,1])\to\mathbb R,\qquad T_{n,x}(f):=(\mathcal L^n f)(x). \] Then \[ |T_{n,x}(f)|\le \Lambda_n(x)\,\|f\|_\infty\quad\text{for all }f\in C([-1,1]), \] and the operator norm satisfies \[ \|T_{n,x}\|=\sup_{\|f\|_\infty\le 1}|T_{n,x}(f)|=\Lambda_n(x). \]

Proof. From the definition, \[ T_{n,x}(f)=\sum_{i=1}^n f(a_i^n)p_i^n(x). \] Taking absolute values and using $|f(a_i^n)|\le\|f\|_\infty$ gives \[ |T_{n,x}(f)|\le \sum_{i=1}^n |f(a_i^n)|\,|p_i^n(x)|\le \|f\|_\infty\sum_{i=1}^n |p_i^n(x)|=\Lambda_n(x)\|f\|_\infty. \] So $\|T_{n,x}\|\le\Lambda_n(x)$.

For the reverse inequality, define a function $g$ on the finite node set by \[ g(a_i^n):=\begin{cases} 1,& p_i^n(x)\ge 0,\\ -1,& p_i^n(x)<0. \end{cases} \] Then $|g(a_i^n)|=1$ for all $i$. Since the node set is finite, we can extend these prescribed values to a continuous function $f\in C([-1,1])$ with $\|f\|_\infty\le 1$ (for instance: sort the nodes in increasing order and define $f$ piecewise linearly so that $f$ takes the value $g(a_i^n)$ at each node; this produces a continuous $f$ with values in $[-1,1]$).

For this $f$, we have \[ T_{n,x}(f)=\sum_{i=1}^n f(a_i^n)p_i^n(x)=\sum_{i=1}^n \operatorname{sgn}(p_i^n(x))\,p_i^n(x)=\sum_{i=1}^n |p_i^n(x)|=\Lambda_n(x). \] Hence $\|T_{n,x}\|\ge \Lambda_n(x)$. Combining with the upper bound yields $\|T_{n,x}\|=\Lambda_n(x)$.

Lemma 671.3 (error bound via best polynomial approximation)

For $f\in C([-1,1])$ define the best uniform approximation error by degree $\le n-1$ polynomials: \[ E_{n-1}(f):=\inf\{\|f-q\|_\infty: q\in\mathbb R[x],\ \deg q\le n-1\}. \] Then for every $x\in[-1,1]$, \[ |(\mathcal L^n f)(x)-f(x)|\le (1+\Lambda_n(x))\,E_{n-1}(f). \]

Proof. Fix any polynomial $q$ with $\deg q\le n-1$. By Lemma 671.1, $\mathcal L^n q=q$. Thus \[ (\mathcal L^n f)(x)-f(x)=(\mathcal L^n (f-q))(x)-(f-q)(x). \] Taking absolute values and applying the triangle inequality, \[ |(\mathcal L^n f)(x)-f(x)|\le |(\mathcal L^n (f-q))(x)|+|(f-q)(x)|. \] By Lemma 671.2 (the bound $|T_{n,x}(h)|\le\Lambda_n(x)\|h\|_\infty$ applied to $h=f-q$), \[ |(\mathcal L^n (f-q))(x)|\le\Lambda_n(x)\,\|f-q\|_\infty. \] Also $|(f-q)(x)|\le\|f-q\|_\infty$. Therefore \[ |(\mathcal L^n f)(x)-f(x)|\le (\Lambda_n(x)+1)\,\|f-q\|_\infty. \] Taking the infimum over all such $q$ gives the claimed inequality.

Proposition 671.4 (uniform boundedness at a fixed point)

Fix a point $x\in[-1,1]$. If $\sup_n \Lambda_n(x)=\infty$, then there exists $f\in C([-1,1])$ such that \[ \sup_n |(\mathcal L^n f)(x)|=\infty. \] In particular, $(\mathcal L^n f)(x)$ does not converge.

Proof. By Lemma 671.2, $\|T_{n,x}\|=\Lambda_n(x)$. Thus $\sup_n\|T_{n,x}\|=\infty$. The Banach space $C([-1,1])$ is complete under $\|\cdot\|_\infty$, and $\{T_{n,x}\}$ is a family of bounded linear functionals on it. By the Uniform Boundedness Principle (Banach--Steinhaus theorem), if $\sup_n\|T_{n,x}\|=\infty$ then there exists $f\in C([-1,1])$ with $\sup_n |T_{n,x}(f)|=\infty$. This is exactly the displayed statement.

5) VERIFICATION

- Lemma 671.1: checked that the argument uses only (i) degree bound $\le n-1$, (ii) $n$ distinct nodes. No hidden assumptions. - Lemma 671.2: the only nontrivial step is extending prescribed values on finitely many points to a continuous function with sup-norm $\le 1$. The explicit piecewise-linear construction works because $[-1,1]$ is an interval and the node set is finite. - Lemma 671.3: uses only Lemma 671.1 and the bound from Lemma 671.2. - Proposition 671.4: assumes the Uniform Boundedness Principle; this is a standard theorem of functional analysis. If one wishes to avoid it, one could reproduce its proof, but I did not.

Boundary cases: for $n=1$, $\Lambda_1(x)=1$ and all inequalities are trivially valid.

6) FINAL

UNRESOLVED

(i) Strongest proved partial result:

For any node choice and any $f\in C([-1,1])$, \[ |(\mathcal L^n f)(x)-f(x)|\le (1+\Lambda_n(x))E_{n-1}(f)\quad\text{for every }x\in[-1,1]. \] Also, for each fixed $x$, if $\sup_n\Lambda_n(x)=\infty$ then there exists $f\in C([-1,1])$ such that $(\mathcal L^n f)(x)$ is unbounded (hence diverges) at that same $x$.

(ii) First gap (crisp):

Given an arbitrary continuous $f$, must there exist a point $x$ with $\limsup_n\Lambda_n(x)=\infty$ such that nevertheless $(\mathcal L^n f)(x)\to f(x)$? I do not have a method to force convergence at a point where $\Lambda_n(x)$ is unbounded.

(iii) Top 3 next moves:

1. Attempt a Baire-category argument on $C([-1,1])$ for fixed nodes, studying the set of $f$ for which $\mathcal L^n f(x)$ converges at a given $x$ versus the growth rate of $\Lambda_n(x)$. 2. For candidate nodes (e.g. Chebyshev-type), try to quantify pointwise growth of $\Lambda_n(x)$ and relate it to typical decay rates of $E_{n-1}(f)$ via explicit modulus of continuity estimates. 3. Computationally: for concrete node systems, numerically approximate the set of $x$ where $\Lambda_n(x)$ is "small" infinitely often, and test pointwise convergence for slowly-approximable continuous functions.

(iv) Minimal counterexample structure:

To refute $Q1$ for a given node system, one would want a continuous $f$ such that for every $x$ with $\limsup_n\Lambda_n(x)=\infty$, the sequence $(\mathcal L^n f)(x)$ fails to converge (ideally is unbounded). Proposition 671.4 produces, for each fixed $x$, some $f_x$ with divergence at that $x$, but I do not know how to construct a single $f$ that forces divergence simultaneously on the full (possibly large) set of divergence points.