1: \begin{abstract}

2: In a $3$-$\mathsf{XOR}$ game $\mathcal{G}$, the verifier samples a challenge $(x,y,z)\sim \mu$

3: where $\mu$ is a probability distribution over $\Sigma\times\Gamma\times\Phi$, 

4: and a map $t\colon \Sigma\times\Gamma\times\Phi\to\mathcal{A}$ for a finite

5: Abelian group $\mathcal{A}$ defining a constraint. 

6: The verifier

7: sends the questions $x$, $y$ and 

8: $z$ to the players Alice, Bob and 

9: Charlie respectively, receives answers $f(x)$, $g(y)$ and $h(z)$ 

10: that are elements in $\mathcal{A}$ 

11: and accepts if $f(x)+g(y)+h(z) = t(x,y,z)$. The value, $\mathsf{val}(\mathcal{G})$, of the game

12: is defined to be the maximum 

13: probability the verifier accepts 

14: over all players' strategies.

15: 

16: We show that if $\mathcal{G}$ is a $3$-$\mathsf{XOR}$

17: game with value strictly less than $1$, whose underlying distribution over

18: questions $\mu$ does not admit Abelian

19: embeddings into $(\mathbb{Z},+)$, then

20: the value of the $n$-fold repetition

21: of $\mathcal{G}$ is exponentially decaying. That is, there exists 

22: $c=c(\mathcal{G})>0$ such that $\mathsf{val}(\mathcal{G}^{\otimes n})\leq 2^{-cn}$. 

23: This extends a previous result of 

24: [Braverman-Khot-Minzer, FOCS 2023] showing 

25: exponential decay for the GHZ game.

26: Our proof combines tools from additive

27: combinatorics and tools from discrete

28: Fourier analysis.

29: \end{abstract}

30: