Pith's one-line read
This paper determines the exact optimal constant and all equality cases for the Poincaré–Wirtinger inequality on the complete graph, for every exponent p in the stated ranges.
desk verdict
Sharp constants on complete graphs are likely correct, but the equality-case classification is false at n=3, p=4, making the claim overbroad; fixable with a stricter monotonicity argument.
read the letter →
A machine-rendered reading of the paper's core claim, the
machinery that carries it, and where it could break.
The reading
The paper solves the extremal problem of comparing a function's deviation from its mean with its total p-variation on the complete graph K_n. For each p in [1, 3+$δ^{1}$_n) ∪ (3+$δ^{2}$_n, ∞), it pins down the sharp multiplicative constant and, more, identifies every function that attains equality. The answer is piecewise: balanced two-valued functions for 1
What carries the argument
The load-bearing object is an auxiliary sharp inequality (Lemma 3) comparing (max f - m)^{p-1} + (m - min f)^{p-1} with the sum over vertices of (max f - f(a_j))^{p-1} + (f(a_j) - min f)^{p-1}. The argument shows that any extremizer of the original quotient satisfies the Euler-Lagrange-type equation (2.3), and that equation is exactly the equality case of this auxiliary inequality; classifying the maximizers of the auxiliary inequality therefore classifies the extremizers of the main problem. The proof of the auxiliary inequality uses symmetrization, convexity and concavity via majorization arguments, and, in the delicate interval 3<p≤3+$δ^{1}$_n, a binomial-series expansion whose coefficients are controlled by a sign-counting rule for polynomial roots.
What would settle it
For n=5 and a fine grid of p in (3, 3+$δ^{1}$_5), compute the maximum of \|f-m_f\|_p^p / \mathrm{Var}_p(f)^p over the one-parameter family f(a_1)=1, f(a_5)=-1, f(a_2)=f(a_3)=f(a_4)=x, x∈[0,1]. The theorem predicts this maximum is exactly 1/($2^{{p-1}}$+3) and is attained at x=0. If any x≠0 yields a strictly larger value, the sharp constant for that p is wrong. A second check: evaluate the power-series inequality (2.14) at x=1/2 for n=5 and p=3+$δ^{1}$_5; if the infinite sum is negative, the auxiliary lemma fails.
On the complete graph with n≥3 vertices, for every exponent p in the stated union of intervals the paper proves that the optimal constant C_{n,p} in \|f-m_f\|_p ≤ C_{n,p} \mathrm{Var}_p(f) is exactly: 2/n for p=1; (⌊n/2⌋^{p-1}+⌈n/2⌉^{p-1})/n^p for 1<p<2; 1/n (an identity) for p=2; 1/($2^{{p-1}}$+n-2) for 2<p≤3+$δ_n^{1}$; and (1+(n-1)^{p-1})/n^p for p≥4 and for p∈(3,4) satisfying the explicit condition (n-2)/n ≥ (1+(n-1)^{p-1})/$n^{{p-1}}$. It also lists the equality cases in each regime: two-valued functions with balanced level sizes for 1<p<2; functions taking values a+c, a, ..., a, a-c (with the interior value exactly the midpoint) for 2<p≤3+$δ_n^{1}$; and single-spike functions for the large-p regime. The proof obtains these constants by showing that every extremizer satisfies a pointwise critical equation that is precisely the equality case of an auxiliary sharp inequality on the extreme deviations.
Load-bearing premise
Everything for 3<p≤3+$δ^{1}$_n rests on the auxiliary inequality (2.5) being exactly sharp, and its proof in that window depends on a sign analysis of a truncated binomial series; if that sign analysis has a gap, the claimed sharp constant in that interval collapses even if the surrounding cases stand.
Editorial extensions
If this is right
Any use of the Poincaré–Wirtinger inequality on K_n can now quote the sharp constant instead of a bound; no improvement is possible within the stated p-ranges.
The equality-case list is complete: for each regime, the only functions attaining the constant are the specified two-valued, three-valued, or spike functions up to translation and scaling.
For p=2 the inequality is an identity with constant 1/n, so K_n is a graph on which every function is an extremizer.
The threshold structure shows a genuine phase transition at p=3: below it the extremizer is a midpoint three-valued function, above it (in the covered range) the extremizer is a single spike, with a small open window between 3+δ^1_n and 3+δ^2_n where the sharp constant is not claimed.
The p=1 endpoint has constant 2/n and its extremizers are exactly two-valued functions with any split, while for 1<p<2 the split must be as balanced as possible.
Reading between the lines
Editorial extensions of the paper, not claims the author makes directly.
The same auxiliary-inequality strategy could be tried on other vertex-transitive graphs; if it works, one would expect the extremizers to organize into the same three families, with crossover exponents set by the graph's diameter or spectral gap rather than by n alone.
The small window (3+δ^1_n, 3+δ^2_n] that the theorem leaves untreated is a natural place to search for a fourth family of maximizers, or for a continuum of extremizers; the paper's own formulas make the gap visible and give a concrete boundary to test.
Because the constants are explicit and the equality cases are finite-dimensional, the result yields a ready-made finite set of test functions for numerical algorithms that estimate Poincaré constants on graphs.