Erdős Problem #671
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: analysis.
Read this status correctly
The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.
The question in plain language
When we approximate a continuous function by a polynomial through chosen sample points, can the interpolation procedure be unstable at a point but still converge there for the function under consideration?
Erdős Problem #671 asks two carefully quantified versions of that question for sequences of Lagrange interpolation nodes in \([-1,1]\). Each row of the node array has \(n\) distinct points \(a_1^n,\ldots,a_n^n\), with Lagrange cardinal polynomials \(p_i^n\). Define
The quantity \(\Lambda_n(x)\) measures how strongly interpolation at \(x\) can amplify small changes to values at the nodes. The problem remains OPEN in the maintained catalogue.
A small example
Take three nodes at \(-1,0,1\). Their cardinal polynomials are
At \(x=1/2\) the values are \(-1/8,3/4,3/8\). They sum to one, yet the sum of their absolute values is
For the quadratic function \(f(x)=x^2\), interpolation is exact because a degree-two polynomial is reproduced by the three-node interpolant. Thus \(L_3f(1/2)=1/4=f(1/2)\) even though \(\Lambda_3(1/2)>1\).
This finite calculation illustrates that numerical instability and exact reproduction of a particular function can coexist. It does not exhibit divergence of \(\Lambda_n\) or solve the infinite-sequence question.
What is known
The first question asks whether one can choose the entire node array so that, for every continuous function \(f\), there is a point \(x\) (allowed to depend on \(f\)) satisfying both
The second asks whether some array can have \(\limsup_n\Lambda_n(x)=\infty\) at every point of the interval, and still, for every continuous function \(f\), have some point \(x\) at which its interpolants converge to \(f(x)\). These two orders of quantifiers must not be interchanged.
The maintained record attributes to Bernstein [Be31] the older theorem that, no matter what nodes are chosen, there is at least one point where the interpolation Lebesgue functions become unbounded along a subsequence. It attributes to Erdős and Vértesi [ErVe80] the existence, for every node array, of a continuous function whose interpolation diverges in magnitude almost everywhere. Neither attribution alone answers the simultaneous existence of a point where a specified function converges and the Lebesgue function diverges.
There is also a useful functional-analytic warning. For any two continuous functions,
If the operator norms \(\Lambda_n(x)\) are unbounded at one fixed \(x\), the uniform boundedness principle gives a continuous function with unbounded interpolation values at that particular point. This does not rule out the problem's possibility of finding a different good point for each function. The quantifier distinction is the heart of the question.
GCL's current state
The authenticated issue-bound #671 source audit contains the exact two requested logical structures, the three-node fixture and the operator inequality. Its disposition is PARTIAL because the original Bernstein, Erdős–Vértesi and historical Erdős proof pages have not been inspected directly.
This is an unaccepted draft, not a resolution of the quantified interpolation questions and not a new functional-analysis theorem.
What remains
For a bounded mathematical successor, fix one explicit triangular array of interpolation nodes and investigate two sets:
The first problem asks whether it is possible to choose nodes with \(D\cap C_f\neq\varnothing\) for every continuous \(f\). The second strengthens the condition to \(D=[-1,1]\) but still permits the convergence point to depend on \(f\).
Any use of Baire category or uniform boundedness must preserve the order of quantifiers. Finite-degree experiments are illustrations and potential checks on an intermediate lemma, never a proof of the desired simultaneous infinite behavior.
Obtain the primary [Er82e], [Er97f], [Be31] and [ErVe80] sources, then finish critical source review, protected publication, deployed mathematical rendering inspection and separate V2 admission.
Sources and claim boundaries
-
Maintained Erdős Problem #671, exact two statements and classical attributions.
-
Authenticated GCL original issue return #671, source audit and finite interpolation check.
-
Bernstein [Be31] and Erdős–Vértesi [ErVe80]: original theorem proofs not independently inspected.
-
Erdős [Er82e], [Er97f], and [Va99, 2.40]: exact historical pages not directly inspected.
Boundary: no solution to the stated quantifier problem, new Banach-space theorem, protected merge, deployed MathJax readback or editorial admission.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1942 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: STATEMENT_CAPTURE_REQUIRED.
- Formal-statement coverage: No statement was confirmed in GCL's pinned formal-provider snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.