diff --git a/src/fri/naive.md b/src/fri/naive.md
index b890043..44d1a6b 100644
--- a/src/fri/naive.md
+++ b/src/fri/naive.md
@@ -111,7 +111,7 @@ $$
 
 
 
-In this manner, the prover can keep calculating for the whole domain. Then they would response the verifier query with the evaluations across layers for checking if an instance $a_i$ corresponding to the query is in the $\mathcal{L}\left(\mathcal{R}_{\text{consistent\_layers }}\right)$
+In this manner, the prover can keep calculating for the whole domain. Then they would response the verifier query with the evaluations across layers for checking if an instance $a_i$ corresponding to the query is in the $\mathcal{L}\left(\mathcal{R}_{\text{consistent-layers }}\right)$
 
 ### Consistency check
 
@@ -150,7 +150,7 @@ $$
 
 Indeed, the value provided by prover, $p_2((2^2)^2)$, is consistent with the recursively accumulated sum from symmetric points on previous layers.
 
-Therefore, $(a_2,w_2)$ is a pair in $\mathcal{R}_{\text {consistent\_layers }}$. So through a single query, this naive FRI convince the verifier that $p_0$ has an expected degree bound.
+Therefore, $(a_2,w_2)$ is a pair in $\mathcal{R}_{\text {consistent-layers }}$. So through a single query, this naive FRI convince the verifier that $p_0$ has an expected degree bound.
 
 For the next, we will see how to deal with a malicious prover by improving this naive version.