where k and m are positive integers. A coefficient is negative when it is less than zero. The exceptional negative coefficients conjecture for D′(2,3,n). The only negative coefficients of
n≥0∑D′(2,3,n)qn
occur at n=10 and n=22. The claim refines the expected sign pattern for a series for which full positivity is not asserted.
References
Primary source
George E. Andrews and Mohamed El Bachraoui, “Certain positive q-series and inequalities for two-color partitions”, arXiv:2507.09276 (2025).
The conjecture remains publicly unproved, although a reader-submitted argument claims to identify the two exceptional negative coefficients.
George E. Andrews and Mohamed El Bachraoui proposed that the series has negative coefficients only at n=10 and n=22. Their preprint appeared in 2025, with a journal version in 2026.
Known results
Andrews and El Bachraoui prove positivity for the related C′(2,3,n) series, but explicitly do not prove the corresponding assertion for D′(2,3,n).
Community submission (unverified)
A submitted proof argues that D′(2,3,n)<0 exactly when n∈{10,22}, with D′(2,3,10)=D′(2,3,22)=−1. It develops generating-function identities, divisor-count formulas, and a bound intended to establish the remaining nonnegativity; no independent verification was found.
Current status (as of August 2026): The conjecture remains open in the published literature; a submitted proof claims a complete resolution, but that argument is unverified.
The factor qj is essential: without it, the infinite sum would not define a formal power series. For a=2j+2, source Lemma 5.3 gives the partial-fraction identity
Indeed, the first two progressions in question correspond to factorizations
2M−1=(2j+3)(2t+1),2M−1=(2j+5)(2t+1),(9)
respectively, while the third corresponds to
2M+3=(2j+5)(2t+1),j,t≥0.(10)
The excluded first factors are 1, and additionally 3 whenever it divides the relevant integer.
Set u0=0. Equations (6) and (8) give the exact integer recurrences
uM=2τ(2M−1)−2−13∣(2M−1)−uM−1(M≥1),(11)
and, writing Fn=[qn]F(q),
Fn=n+4−τ(2n+9)+13∣n−un+3(n≥0).(12)
To bound uM, let χ4 be the nonprincipal character modulo 4, and define
S(x)=d≤x∑χ4(d),A(X)=d≤X∑χ4(d)τ(d).(13)
Because S(x)∈{0,1} and χ4 is completely multiplicative, the Dirichlet hyperbola identity gives
A(X)=2a≤X∑χ4(a)S(aX)−S(X)2.(14)
Only odd a contribute. Consequently,
∣A(X)∣≤2⌈2⌊X⌋⌉+1≤X+2.(15)
Since
(−1)t=−χ4(2t−1),(16)
the alternating sum of the divisor terms in (11) is, up to sign, 2A(2M−1). The constant terms contribute at most 2 in absolute value, and the terms with 3∣(2t−1) alternate because their indices differ by 3, so they contribute at most 1. Therefore
∣uM∣≤22M−1+7(M≥1).(17)
Moreover, τ(N)≤2N. Combining (8), (12), and (17) yields
Exactly eight of the sixteen odd residue pairs modulo 8 satisfy this congruence, for either parity of n. Partition the first quadrant into half-open 8×8 squares. Each contains exactly eight admissible points, and any square containing a point of norm at most R=8n+2 lies inside the quarter disk of radius R+82. Comparing their areas and using π<22/7 gives
Indeed, bn is one quarter of the number of signed representations of 8n+2 as a sum of two squares, and its odd divisors are precisely the divisors of 4n+1. Equation (19) now gives
H0=b0,H1=b1,Hn=bn+Hn−2(n≥2).(30)
Thus (11), (12), (29), and (30) determine every coefficient using only ordinary divisor counts and the character χ4. Put
dn=Fn−Hn.(31)
Exact evaluation over the entire remaining interval gives the following finite certificate; the final column provides an independent block-total check.
The intervals in (32) partition all integers from 0 through 3286. The only negative entries occur at n=10 and n=22, and in both cases the exact coefficient is −1. Combining this complete finite certificate with the unconditional infinite-range inequality (28) proves (1).