Let k k k be a positive integer, let r r r be a positive fundamental discriminant, and let χ r \chi_r χ r be the primitive quadratic character modulo r r r . Define
T d k ; x S ( r ) : = ∑ n ≤ x ( n , r ) = 1 d k ( n ) χ r ( n ) . \mathcal{T}^S_{d_k;x}(r):=\sum_{\substack{n\leq x\\(n,r)=1}}d_k(n)\chi_r(n). T d k ; x S ( r ) := n ≤ x ( n , r ) = 1 ∑ d k ( n ) χ r ( n ) .
For y > 0 y>0 y > 0 , define the variance over positive fundamental discriminants in ( y , 2 y ] (y,2y] ( y , 2 y ] by
Var r ∈ ( y , 2 y ] ( T d k ; x S ) : = E y < r ≤ 2 y ∗ ( T d k ; x S ( r ) ) 2 . \operatorname{Var}_{r\in(y,2y]}(\mathcal{T}^S_{d_k;x}):=\mathbb E^*_{y<r\leq 2y}\bigl(\mathcal{T}^S_{d_k;x}(r)\bigr)^2. Var r ∈ ( y , 2 y ] ( T d k ; x S ) := E y < r ≤ 2 y ∗ ( T d k ; x S ( r ) ) 2 .
Here E ∗ \mathbb E^* E ∗ denotes expectation over positive fundamental discriminants. Kuperberg–Lalin's symplectic variance conjecture. If x 1 / k + ε ≤ y x^{1/k+\varepsilon}\leq y x 1/ k + ε ≤ y , then
Var r ∈ ( y , 2 y ] ( T d k ; x S ) ∼ a k S ( T ) x γ d k , 2 S ( log x log y ) ( log y ) 2 k 2 + k − 2 , \operatorname{Var}_{r\in(y,2y]}(\mathcal{T}^S_{d_k;x})\sim a_k^S(\mathcal T)x\gamma_{d_k,2}^S\left(\frac{\log x}{\log y}\right)(\log y)^{2k^2+k-2}, Var r ∈ ( y , 2 y ] ( T d k ; x S ) ∼ a k S ( T ) x γ d k , 2 S ( log y log x ) ( log y ) 2 k 2 + k − 2 ,
where
a k S ( T ) = 2 ∏ p ( 1 − 1 p ) k ( 2 k + 1 ) ( 1 p + 1 ( 1 + p 2 ( ( 1 + 1 p ) − 2 k + ( 1 − 1 p ) − 2 k ) ) ) , a_k^S(\mathcal T)=2\prod_p\left(1-\frac1p\right)^{k(2k+1)}\left(\frac1{p+1}\left(1+\frac p2\left(\left(1+\frac1{\sqrt p}\right)^{-2k}+\left(1-\frac1{\sqrt p}\right)^{-2k}\right)\right)\right), a k S ( T ) = 2 p ∏ ( 1 − p 1 ) k ( 2 k + 1 ) ( p + 1 1 ( 1 + 2 p ( ( 1 + p 1 ) − 2 k + ( 1 − p 1 ) − 2 k ) ) ) ,
and γ d k , 2 S ( c ) \gamma_{d_k,2}^S(c) γ d k , 2 S ( c ) is the piecewise polynomial of degree 2 k 2 + k − 2 2k^2+k-2 2 k 2 + k − 2 defined as the leading coefficient in N N N of
I d k , 2 S ( c 2 N ; N ) : = ∫ Sp ( 2 N ) ∣ ∑ j 1 + ⋯ + j k = c 2 N 0 ≤ j 1 , … , j k ≤ 2 N Sc j 1 ( U ) ⋯ Sc j k ( U ) ∣ 2 d U . I_{d_k,2}^S(c2N;N):=\int_{\operatorname{Sp}(2N)}\left|\sum_{\substack{j_1+\cdots+j_k=c2N\\0\leq j_1,\ldots,j_k\leq 2N}}\operatorname{Sc}_{j_1}(U)\cdots\operatorname{Sc}_{j_k}(U)\right|^2\,\mathrm dU. I d k , 2 S ( c 2 N ; N ) := ∫ Sp ( 2 N ) j 1 + ⋯ + j k = c 2 N 0 ≤ j 1 , … , j k ≤ 2 N ∑ Sc j 1 ( U ) ⋯ Sc j k ( U ) 2 d U .
The conjecture is motivated by an analogous function-field result and incorporates the arithmetic factor a k S ( T ) a_k^S(\mathcal T) a k S ( T ) . The source states that it is proved when y ≥ x 2 ( log x ) C y\geq x^2(\log x)^C y ≥ x 2 ( log x ) C for a sufficiently large constant C C C , leaving the stated range beyond that regime open.