-
Notifications
You must be signed in to change notification settings - Fork 6
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
feat(basefold): add list decoding soundness details
- Loading branch information
1 parent
702703c
commit 2b82f9e
Showing
1 changed file
with
296 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,296 @@ | ||
# 笔记:Basefold 在 List Decoding 下的 Soundness 证明 | ||
|
||
- Jade Xie <[email protected]> | ||
- Yu Guo <[email protected]> | ||
|
||
在上一篇文章《Basefold 在 List Decoding 下的 Soundness 证明概览》中,梳理了 [H24] 论文中 soundness 证明的思路,本篇文章将沿着这个思路深入论文中的证明细节,主要是 [H24, Lemma 1] 的证明,其证明了 Basefold 协议在 commit 阶段的 soundness error。 | ||
|
||
**Lemma 1** [H24, Lemma 1] (Soundness commit phase). Take a proximity parameter $\theta=1-\left(1 + \frac{1}{2 \cdot m}\right) \cdot \sqrt{\rho}$, with $m\geq3$. Suppose that a (possibly computationally unbounded) algorithm $P^*$ succeeds the commitment phase with $r\geq0$ rounds with probability larger than | ||
|
||
$$ | ||
\varepsilon_C=\varepsilon_0+\varepsilon_1+\ldots+\varepsilon_r, | ||
$$ | ||
|
||
where $\varepsilon_0=\varepsilon(\mathcal{C}_i,M,\theta)$ is the soundness error from Theorem 3, and | ||
|
||
$$ | ||
\varepsilon_i:=\varepsilon(\mathcal{C}_i,1,B_i,\theta)+\frac{1}{|F|}, | ||
$$ | ||
|
||
with $\varepsilon(\mathcal{C}_i,1,B_i,\theta)$ being the soundness error from Theorem 4, where $B_i=\frac{|D|}{|D_i|}=2^i$. Then $(g_0,\ldots,g_M)$ belongs to $\mathcal{R}$. | ||
|
||
引理中提到的 [H24, Theorem 3] 就是在 list decoding 下针对 subcode 的 correlated agreement 定理,而 [H24, Theorem 4] 就是 [H24, Theorem 3] 的 weighted 版本。 | ||
|
||
关系 $\mathcal{R}$ 表示的含义是能得出 $P^*$ 没有作恶,说明其承诺的多项式 $(g_{0}, \ldots, g_{M})$ 既离对应的编码空间距离不超过 $\theta$ ,同时也满足在查询点 $\vec{\omega} = (\omega_1, \ldots, \omega_n)$ 的值与承诺的值 $v_0, \ldots, v_M$ 是一致的,即 | ||
|
||
$$ | ||
\mathcal{R}=\left\{\begin{array}{c} | ||
\exists p_0, \ldots, p_M \in \mathscr{F}[X]^{<2^n} \text { s.t. } \\ | ||
\left(g_0, \ldots, g_M\right): d\left(\left(g_0, \ldots, g_M\right),\left(p_0, \ldots, p_M\right)\right)<\theta \\ | ||
\wedge \bigwedge_{k=0}^M P_k\left(\omega_1, \ldots, \omega_n\right)=v_k | ||
\end{array}\right\}. | ||
$$ | ||
Lemma 1 说明的就是如果 $P^*$ 在 commit 阶段成功的概率超过了 $\varepsilon_C$ ,那么我们能相信 $P^*$ 没有作弊,其声称的关系 $\mathcal{R}$ 也是成立的。 | ||
|
||
在这里,还需要用数学语言去定义 $P^*$ 在 commit 阶段的第 $0 \le r \le n$ 轮能成功的含义,这就是 [H24] 论文中给出的 $\alpha$ -good 的概念。从协议本身理解,$P^*$ 能成功,意味着 verifier 拿到 $P^*$ 发送过来的 $f_0, \Lambda_0, f_1, \Lambda_1, f_2, \Lambda_2, \ldots, \Lambda_{r-1}, f_r$ ,然后进行检查,一是进行 sumcheck 的检查,另一个是在 $D_0$ 中随机选取 $x$ ,FRI 的折叠是正确的。首先这里的参数 $\alpha = 1 - \theta \in (0,1)$ ,即 | ||
|
||
$$ | ||
\alpha = \left(1 + \frac{1}{2 \cdot m}\right) \cdot \sqrt{\rho} | ||
$$ | ||
用 $\mathcal{F}_i$ 表示和 Reed-Solomon 编码 $\mathcal{C}_i = \mathrm{RS}_{2^{n-i}}[F, D_i]$ 相对应的多项式空间,其中 $D_i$ 就是用映射 $\pi$ 对 $D$ 作用 $i$ 次,即 $D_i = \pi^i(D), i = 0, \ldots, n$ 。因此与 $\mathcal{C}_i' \subseteq \mathcal{C}_i$ 相对应的多项式子空间定义为 | ||
|
||
$$ | ||
\mathcal{F}_i' = \left\{p(X) \in \mathcal{F}_i: P(\omega_{i + 1}, \ldots, \omega_n) = 0 \right\}. | ||
$$ | ||
|
||
1. sumcheck 检查正确。意味着存在 $p_r(X) \in \mathcal{F}_r$ ,其对应的多元多项式为 $P_r$ 满足 | ||
|
||
$$ | ||
L((\omega_1, \ldots, \omega_r), (\lambda_1, \ldots, \lambda_r)) \cdot P_r(\omega_1, \ldots, \omega_n) = q_{r-1}(\lambda_r) | ||
$$ | ||
|
||
根据 $q_i(X)$ 与 $\Lambda_i(X)$ 之间的关系,可以得到 $P_r$ 需要满足 | ||
|
||
$$ | ||
\begin{aligned} | ||
L((\omega_1, \ldots, \omega_r), (\lambda_1, \ldots, \lambda_r)) \cdot & P_r(\omega_{r + 1}, \ldots, \omega_n) = q_{r-1}(\lambda_r) \\ | ||
& = L((\omega_1, \ldots, \omega_r), (\lambda_1, \ldots, \lambda_r)) \cdot \Lambda_{r - 1}(\lambda_r) | ||
\end{aligned} \tag{1} | ||
$$ | ||
|
||
2. 折叠正确。需要满足 | ||
|
||
$$ | ||
\left| \left\{ x \in D_0 : \quad \begin{array}{c} | ||
(f_0, \ldots, f_r) \text{ satisfy all folding checks along } x \\ | ||
\wedge f_r(\pi^r(x)) = p_r(\pi^r(x)) | ||
\end{array}\right\}\right| \ge \alpha \cdot |D_0| \tag{2} | ||
$$ | ||
这里只有当在 $D_0$ 中满足 folding check 的 $x$ 的比例大于 $\alpha$ ,经过 $\pi^r$ 映射,到最后 verifier 才会通过。 | ||
|
||
当满足 1 和 2 两个条件时,就说这样的 $(f_0, \Lambda_0, f_1, \Lambda_1, f_2, \Lambda_2, \ldots, \Lambda_{r-1}, f_r)$ 对于 $(\lambda_0, \ldots, \lambda_r)$ 来说 $\alpha$ -good 的。 | ||
|
||
## Lemma 1 证明 | ||
|
||
Lemma 1 的证明采用的是数学归纳法,先证明当 $r = 0$ 时结论是成立的,这里用到了 [H24, Therorem 3]。接着假设 Lemma 1 在 $0 \le r < n$ 时成立,证明 Lemma 1 在 $r + 1$ 时结论也成立,在这个过程中就用到了带权重的 [H24, Theorem 4] ,其证明思路与上篇文章介绍的思路类似。例如在第 $r + 1$ 轮,用随机数 $\lambda_{r + 1}$ 折叠之后得到 $f_{r + 1}$ 满足的条件入手,其离对应的编码空间距离比较近,并满足 sumcheck 约束,先推导出对应的 $f_{r + 1}'$ 满足一些条件,这样就能使用针对 subcode 的 correlated agreement 定理了。应用定理的结论,进而得到在折叠之前的 $f_{r,0}$ 与 $f_{r,1}$ 满足的性质,以此再得出 $f_r$ 满足的性质。此时应用归纳假设,能得到在第 $r$ 轮满足引理的条件,从而得出在第 $r$ 轮的结论成立,也就证明了在第 $r + 1$ 轮引理成立。 | ||
|
||
证明:首先证明当 $r = 0$ 时引理是成立的。已知的条件是 $P^*$ 在 commit 阶段成功的概率大于 $\varepsilon(\mathcal{C}_0,M,\theta)$ ,想证明得到的结论是 $(g_1, \ldots, g_M) \in \mathcal{R}$ 。根据条件以及 $\alpha$ -good 的定义,可以得到以大于 $\varepsilon(\mathcal{C}_0,M,\theta)$ 的概率 $P^*$ 提供的 $f_0$ 对 $\lambda_0$ 来说是 $\alpha$ -good 的,那么对于考虑折叠之前的多项式 $g_k' = g_k - v_k$ ,距离对应的 subcode $\mathcal{C}_0' \subseteq \mathcal{C}_0$ 不超过 $\theta$ (也就说明一致的地方大于 $\alpha$ )的概率为 | ||
|
||
$$ | ||
\Pr \left[ \lambda_0: \exists p_0' \in \mathcal{F}_0' \text{ s.t. } \mathrm{agree} \left( \sum_{k = 0}^{M} g_k' \cdot \lambda_0^k, p_0'(X) \right) \ge \alpha \right] > \varepsilon(\mathcal{C}_0,M,\theta) | ||
$$ | ||
|
||
这里考虑的是多项式 $g_k' = g_k - v_k$ 而不是 $g_k$ 的目的是,能让我们的分析进入线性子码 $\mathcal{C}_0'$ 的范围内,这样我们就能用 [H24, Theorem 3] ,得到存在多项式 | ||
|
||
$$ | ||
p_0'(X), \ldots, p_M'(X) \in \mathcal{F}_0' | ||
$$ | ||
|
||
以及存在集合 $D_0' \subseteq D$ ,满足 | ||
|
||
1. $|D_0'|/|D| \ge \alpha$ | ||
2. $p_k'(X)|_{D_0'} = g_k'(X)|_{D_0'}$ | ||
|
||
现在找到了多项式 $p_0'(X), \ldots, p_M'(X)$ ,那么对于多项式 | ||
|
||
$$ | ||
p_0'(X) + v_0, \ldots, p_M'(X) + v_M \in \mathcal{F}_0 | ||
$$ | ||
|
||
就满足 | ||
|
||
$$ | ||
(p_k'(X) + v_k)|_{D_0'} = (g_k'(X) + v_k)|_{D_0'} = g_k(X)|_{D_0'} \quad 0 \le k \le M | ||
$$ | ||
|
||
$p_0'(X) + v_0$ 对应的多元线性多项式 $P_k \in F[X_1, \ldots, X_n]$ 也满足 $P_k(\vec{\omega}) = v_k$ ,因此 $(g_1, \ldots, g_M) \in \mathcal{R}$ 。 | ||
|
||
现在假设引理在 $0 \le r < n$ 时是成立的,想证明在 $r + 1$ 时引理依然成立。根据引理的条件,在第 $r + 1$ 轮,$P^*$ 在 commit 阶段成功的概率超过 $(\varepsilon_0 + \varepsilon_1 + \ldots + \varepsilon_r) + \varepsilon_{r + 1}$ 。记 $\mathrm{tr}_r = (\lambda_0, f_0, \Lambda_0, \ldots, \lambda_r, f_r, \Lambda_r)$ 组成的集合为 $\mathfrak{T}$ ,因此在 | ||
|
||
$$ | ||
\operatorname{Pr}[\mathfrak{T}]>\varepsilon_0+\ldots+\varepsilon_r | ||
$$ | ||
|
||
的条件下,$P^*$ 成功的概率大于 $\varepsilon_{r + 1}$ ,即 | ||
|
||
$$ | ||
\Pr \left[ \lambda_{r+1}: | ||
\begin{array}{c} | ||
\exists f_{r + 1} \text{ s.t. } (\lambda_0, f_0, \Lambda_0, \ldots, \lambda_r, f_r, \Lambda_r, f_{r + 1}) \\ | ||
\text{is $\alpha$-good for } (\lambda_0, \ldots, \lambda_{r + 1}) | ||
\end{array} | ||
\right] > \varepsilon_{r + 1} | ||
$$ | ||
|
||
由 $\alpha$ - good 的定义可以得到,对于满足 $\alpha$ -good 的 $\lambda_{r + 1}$ ,存在一个满足 sumcheck 约束的多项式 $p_{r + 1} \in \mathcal{F}_{r + 1}$ ,使得 | ||
|
||
$$ | ||
\mathrm{agree}_{\nu_r}((1 - \lambda_{r + 1}) \cdot f_{r,0} + \lambda_{r + 1} \cdot f_{r,1}, p_{r + 1}) \ge \alpha \tag{3} | ||
$$ | ||
|
||
这里的 $\nu_r$ 是一个子概率测度,其 density 函数定义为,对 $y \in D_{r + 1}$ 有 | ||
|
||
$$ | ||
\delta_r(y) : = \frac{|\{x \in \pi^{-(r + 1)}(y): (f_0, \ldots, f_r) \text{ satisfies all folding checks along } x \}|}{|\pi^{-(r+1)}(y)|} | ||
$$ | ||
|
||
这里解释下式 $(3)$ 表示的实质上就是 $\alpha$ -good 定义中的式 $(2)$ 。根据 $\mathrm{agree}$ 函数的定义,式 $(3)$ 等价于 | ||
|
||
$$ | ||
\frac{\nu_r(\{ y \in D_{r + 1}: ((1 - \lambda_{r + 1}) \cdot f_{r,0} + \lambda_{r + 1} \cdot f_{r,1})(y) = p_{r + 1}(y)\})}{|D_{r + 1}|} \ge \alpha | ||
$$ | ||
|
||
先将在 $D_{r + 1}$ 中满足折叠关系的 $y$ 组成一个集合,记为 $S_{r + 1}$ ,再用 $\nu_r$ 函数对这个集合进行计算。 | ||
|
||
$$ | ||
\begin{aligned} | ||
\nu_r (S_{r + 1}) & = \sum_{y \in S_{r + 1}} \delta_r(y) \\ | ||
& = \sum_{y \in S_{r + 1}} \frac{|\{x \in \pi^{-(r + 1)}(y): (f_0, \ldots, f_r) \text{ satisfies all folding checks along } x \}|}{|\pi^{-(r+1)}(y)|} \\ | ||
& = \sum_{y \in S_{r + 1}} \frac{|\{x \in \pi^{-(r + 1)}(y): (f_0, \ldots, f_r) \text{ satisfies all folding checks along } x \}|}{2^{r + 1}} \\ | ||
& := \sum_{y \in S_{r + 1}} \frac{|S_{y,0}|}{2^{r + 1}} \\ | ||
& = \frac{\sum_{y \in S_{r + 1}} |S_{y,0}|}{2^{r + 1}} | ||
\end{aligned} | ||
$$ | ||
|
||
因此 | ||
|
||
$$ | ||
\begin{aligned} | ||
\mathrm{agree}_{\nu_r}((1 - \lambda_{r + 1}) \cdot f_{r,0} + \lambda_{r + 1} \cdot f_{r,1}, p_{r + 1}) & = \frac{\nu_r(S_{r + 1})}{|D_{r + 1}|} \\ | ||
& = \frac{\sum_{y \in S_{r + 1}} |S_{y,0}|}{2^{r + 1} \cdot |D_{r + 1}|} \\ | ||
& = \frac{\sum_{y \in S_{r + 1}} |S_{y,0}|}{|D_{0}|} | ||
\end{aligned} | ||
$$ | ||
|
||
上式中分子 $\sum_{y \in S_{r + 1}} |S_{y,0}|$ 表示的含义正是在 $D_0$ 中满足第 $r + 1$ 次折叠正确,同时 $(f_0, \ldots, f_r)$ 折叠检查也是正确的。$(3)$ 式就变为 | ||
|
||
$$ | ||
\sum_{y \in S_{r + 1}} |S_{y,0}| \ge \alpha \cdot |D_{0}| | ||
$$ | ||
|
||
这与 $\alpha$ -good 定义中式 $(2)$ 是完全一致的。接下来根据在上篇文章中介绍的 soundness 证明思路,由于 $p_{r+1}(X)$ 对应的多元线性多项式 $P_{r+1}$ 满足 sumcheck 约束,因此满足 | ||
|
||
$$ | ||
\begin{aligned} | ||
L((\omega_1, \ldots, \omega_{r+1}), (\lambda_1, \ldots, \lambda_{r + 1})) \cdot & P_{r+1}(\omega_{r + 2}, \ldots, \omega_n) = q_{r}(\lambda_{r + 1}) \\ | ||
& = L((\omega_1, \ldots, \omega_{r + 1}), (\lambda_1, \ldots, \lambda_{r + 1})) \cdot \Lambda_{r}(\lambda_{r + 1}) | ||
\end{aligned} | ||
$$ | ||
|
||
推出 | ||
|
||
$$ | ||
\begin{aligned} | ||
L((\omega_1, \ldots, \omega_{r}), (\lambda_1, \ldots, \lambda_{r})) \cdot L(\omega_{r+1}, \lambda_{r + 1}) \cdot & P_{r+1}(\omega_{r + 2}, \ldots, \omega_n) \\ | ||
& = L((\omega_1, \ldots, \omega_{r}), (\lambda_1, \ldots, \lambda_{r})) \cdot L(\omega_{r+1}, \lambda_{r + 1}) \cdot \Lambda_{r}(\lambda_{r + 1}) | ||
\end{aligned} | ||
$$ | ||
|
||
对于 $\lambda_{r + 1}$ 的选择,有 $1/|F|$ 的概率使得 $L(\omega_{r+1}, \lambda_{r + 1}) = 0$ ,得出上式成立。因此除了 $1/|F|$ 的概率,依然有超过 | ||
|
||
$$ | ||
\varepsilon_{r + 1} - \frac{1}{|F|} = \varepsilon(\mathcal{C}_{i + 1}, 1, B_{r + 1}, \theta) | ||
$$ | ||
|
||
的概率,使得多项式 $p_{r+1}' = p_{r + 1} - \Lambda_r(\lambda_{r + 1}) \in \mathcal{F}_{r + 1}'$ ,以及 $f_{r,0}' = f_{r,0} - \Lambda_r(0)$ ,$f_{r,1}' = f_{r,1} - \Lambda_r(1)$ 满足 | ||
|
||
$$ | ||
\mathrm{agree}_{\nu_r}((1 - \lambda_{r + 1}) \cdot f_{r,0}' + \lambda_{r + 1} \cdot f_{r,1}', p_{r + 1}') \ge \alpha | ||
$$ | ||
|
||
上面满足的条件可以写为 | ||
|
||
$$ | ||
\begin{aligned} | ||
\Pr \left[ \lambda_{r+1}: \quad | ||
\begin{array}{c} | ||
\exists p_{r + 1}' \in \mathcal{F}_{r+1}' \text{ s.t. } \\ | ||
\mathrm{agree}_{\nu_r}((1 - \lambda_{r + 1}) \cdot f_{r,0}' + \lambda_{r + 1} \cdot f_{r,1}', p_{r + 1}') \ge \alpha | ||
\end{array} | ||
\right] > \varepsilon(\mathcal{C}_{i + 1}, 1, B_{r + 1}, \theta) | ||
\end{aligned} | ||
$$ | ||
|
||
这也就满足了 [H24, Theorem 4] 带权重的 correlated agreement 定理的条件,因此可以得到存在多项式 $p_{r,0}'(X), p_{r,1}'(X) \in \mathcal{F}_{r+1}'$ ,以及集合 $A_{r + 1} \subseteq D_{r+1}$ 满足: | ||
1. $\nu_r(A_{r+1}) \ge 1 - \theta$ | ||
2. $p_{r,0}'(X)|_{A_{r+1}} = f_{r,0}'(X)|_{A_{r+1}}$ , $p_{r,1}'(X)|_{A_{r+1}} = f_{r,1}'(X)|_{A_{r+1}}$ | ||
|
||
现在已经找到了多项式 $p_{r,0}'(X), p_{r,1}'(X)$ ,因此存在多项式 | ||
|
||
$$ | ||
p_{r,0}(X) = p_{r,0}'(X) + \Lambda_r(0), \quad p_{r,1}(X) = p_{r,1}'(X) + \Lambda_r(1) \in \mathcal{F}_{r+1} | ||
$$ | ||
|
||
而 | ||
|
||
$$ | ||
f_{r,0}(X) = f_{r,0}'(X) + \Lambda_r(0), \quad f_{r,1}(X) = f_{r,1}'(X) + \Lambda_r(1) | ||
$$ | ||
|
||
根据 correlated agreement 给出的结论 2 ,可以得到 | ||
|
||
$$ | ||
p_{r,0}(X)|_{A_{r+1}} = f_{r,0}(X)|_{A_{r+1}}, \quad p_{r,1}(X)|_{A_{r+1}} = f_{r,1}(X)|_{A_{r+1}} | ||
$$ | ||
|
||
对于 $p_{r,0}(X), p_{r,1}(X)$ 相对应的多元线性多项式 $P_{r,0}$ 以及 $P_{r,1}$ ,根据 $\mathcal{F}_{r}'$ 的定义,可以得到 | ||
|
||
$$ | ||
\begin{aligned} | ||
P_{r,0}(\omega_{r+2}, \ldots, \omega_{n}) = \Lambda_r(0) \\ | ||
P_{r,1}(\omega_{r+2}, \ldots, \omega_{n}) = \Lambda_r(1) | ||
\end{aligned} | ||
$$ | ||
|
||
将集合 $A_{r+1}$ 中的点通过 $\pi$ 的逆映射得到 $A_r = \pi^{-1}(A_{r+1}) \subseteq D_r$ ,在这些点一定满足 $f_r$ 和 | ||
|
||
$$ | ||
p_r(X) = p_{r,0}(X^2) + X \cdot p_{r,1}(X^2) \in \mathcal{F}_{r} | ||
$$ | ||
|
||
是一致的。对于与 $p_r(X)$ 相对应的多元线性多项式 $P_r$ ,其满足 | ||
|
||
$$ | ||
\begin{aligned} | ||
P_r(\omega_{r + 1}, \omega_{r + 2}, \ldots, \omega_n) & = (1 - \omega_{r + 1}) \cdot P_{r,0}(\omega_{r+2}, \ldots, \omega_{n}) + \omega_{r + 1} \cdot P_{r,1}(\omega_{r+2}, \ldots, \omega_{n}) \\ | ||
& = (1 - \omega_{r + 1}) \cdot \Lambda_r(0) + \omega_{r + 1} \cdot \Lambda_r(1) \\ | ||
& = L(\omega_{r + 1}, 0) \cdot \Lambda_r(0) + L(\omega_{r + 1}, 1) \cdot \Lambda_r(1) | ||
\end{aligned} | ||
$$ | ||
|
||
由此可以得到在第 $r$ 轮的 sumcheck 是满足的: | ||
|
||
$$ | ||
\begin{aligned} | ||
L(\omega_{1}, \ldots, \omega_{r}, \lambda_1, \ldots, \lambda_r) &\cdot P_r(\omega_{r + 1}, \omega_{r + 2}, \ldots, \omega_n) \\ | ||
& = L(\omega_{1}, \ldots, \omega_{r}, \lambda_1, \ldots, \lambda_r) \cdot L(\omega_{r + 1}, 0) \cdot \Lambda_r(0) \\ | ||
& \quad + L(\omega_{1}, \ldots, \omega_{r}, \lambda_1, \ldots, \lambda_r) \cdot L(\omega_{r + 1}, 1) \cdot \Lambda_r(1) \\ | ||
& = q_r(0) + q_r(1) \\ | ||
& = q_{r - 1}(\lambda_r) | ||
\end{aligned} | ||
$$ | ||
|
||
现在得到了在第 $r$ 轮的 sumcheck 是满足的,接下来需要考虑折叠关系是否满足。考虑 $x \in \pi^{-1}(A_r)$ ,有 | ||
|
||
$$ | ||
\begin{aligned} | ||
& \frac{|\{x \in \pi^{-r}(A_r): \text{all folding checks hold for } f_0, \ldots, f_r \}|}{|D_0|} \\ | ||
& = \frac{1}{|D_0|} \cdot \sum_{y \in A_{r+1}} \delta(y) \cdot |\pi^{-(r+1)}(y)| \\ | ||
& = \frac{2^{r + 1}}{|D_0|} \cdot \sum_{y \in A_{r+1}} \delta(y) \\ | ||
& = \frac{1}{|D_{r + 1}|} \cdot \sum_{y \in A_{r+1}} \delta(y) \\ | ||
& = \nu_r(A_{r+1}) | ||
\end{aligned} | ||
$$ | ||
|
||
前面通过 correlated agreement 定理已经得到 $\nu_r(A_{r+1}) \ge \alpha$ ,因此在 $D_0$ 中的 $x$ 能满足 folding check 的比例超过 $\alpha$ 。综合在第 $r$ 轮的 sumcheck 约束以及折叠关系,得到 $(f_0, \Lambda_0, \ldots, f_r, \Lambda_r)$ 对于 $(\lambda_0, \ldots, \lambda_r)$ 是 $\alpha$ -good 的。由于产生这样的 trace 的集合的概率 | ||
|
||
$$ | ||
\operatorname{Pr}[\mathfrak{T}]>\varepsilon_0+\ldots+\varepsilon_r | ||
$$ | ||
|
||
因此其满足引理的条件,由归纳假设,在第 $r$ 轮引理成立,因此可以得到结论,$(g_0, \ldots, g_M) \in \mathcal{R}$ ,至此就证明了在第 $r + 1$ 轮引理也是成立的。从而得证引理成立。<span style="float: right;"> $\Box$ </span> | ||
|
||
## References | ||
|
||
- [H24] Ulrich Haböck. "Basefold in the List Decoding Regime." _Cryptology ePrint Archive_(2024). |