diff --git a/docs/attachments/hash_ivc_folding_scheme.png b/docs/attachments/hash_ivc_folding_scheme.png
new file mode 100644
index 0000000..212c599
Binary files /dev/null and b/docs/attachments/hash_ivc_folding_scheme.png differ
diff --git a/docs/attachments/simple_ivc_folding_scheme.png b/docs/attachments/simple_ivc_folding_scheme.png
new file mode 100644
index 0000000..b3dc25d
Binary files /dev/null and b/docs/attachments/simple_ivc_folding_scheme.png differ
diff --git a/docs/attachments/zk_snark_ivc_folding_scheme.png b/docs/attachments/zk_snark_ivc_folding_scheme.png
new file mode 100644
index 0000000..a921509
Binary files /dev/null and b/docs/attachments/zk_snark_ivc_folding_scheme.png differ
diff --git a/docs/nova.md b/docs/nova.md
index f6d1a9b..eb814c2 100644
--- a/docs/nova.md
+++ b/docs/nova.md
@@ -76,3 +76,74 @@ denote a [random oracle](../terms/random_oracle_model.md), then
 the Prover (P) can get random challenge $r$ via:
 
 $r \leftarrow \rho(vk, u_1, u_2, \bar T)$.
+
+## Constructing IVC from a Folding Scheme
+
+### Simple IVC Folding Scheme
+
+The simple idea is: at step $i$ of the [IVC](../terms/recursive_snark.md) process:
+
+- Nova’s prover proves that the step $i$ was computed correctly.
+- Nova treats the computation at step $i-1$ as an R1CS instance and folds it into a running relaxed R1CS instance.
+
+So, each step computes a step of the incremental computation and includes a circuit for the computation of the verifier
+in the non-interactive folding scheme for relaxed R1CS.
+
+In this construction, the prover uses an augmented function $F'$. This function not only invokes $F$ but also performs
+additional bookkeeping to fold proofs of prior invocations of itself.
+
+$F'$ takes as non-deterministic advice two committed relaxed R1CS instances $u_i$ and $U_i$, where:
+
+- $u_i$ represents the correct execution of invocation $i$ of $F'$:  $F'(z_{i-1}) = z_i$. This is the **fresh instance**.
+- $U_i$ represents the correct execution of invocations $1,...,i-1$ of $F'$: $F'^{(i -1)}(z_0) = z_{i-1}$.
+  This is the **running instance**.
+
+$F'$ executes 2 tasks:
+
+- Executes a step of the incremental computation: $F'(z_i) = z_{i + 1}$.
+- Invokes the verifier of non-interactive folding scheme to fold the task of checking $u_i$ and $U_i$ into the task of checking
+- a single instance $U_{i+1}$.
+
+The IVC prover then computes a new instance $u_{i+1}$,which is used to attest that $z_{i + 1} = F'(z_i)$, and gets $U_{i+1}$ as
+the result of folding $u_i$ and $U_i$.
+
+Here is the depiction:
+
+![Simple IVC Folding Scheme](attachments/simple_ivc_folding_scheme.png)
+
+However, the instance $u_{i + 1}$ contains $U_{i + 1}.x$, and $F'$ must hold $u_{i+1}.x$ into $U_{i+1}.x$ in the next iteration,
+so the size of $u_{i+1}$ grows linearly. To avoid that, we replace it with a hash function of its public IO. The next invocation
+of $F'$ will takes the preimage of this hash via the proof.
+
+In particular, the public IO of $u_i$ and $u_{i+1}$ are:
+
+- $u_i.x = hash(i, z_0, z_i, U_i)$
+- $u_{i+1}.x = hash(i, z_0, z_{i+1}, U_{i+1})$
+
+The verifier checks that $u_i.x \stackrel{?}{=} hash(i, z_0, z_i, U_i)$, and the prover now needs to send proof:  
+$((U_{i+1}, W_{i+1}), (u_{i+1}, w_{i+1}))$.
+
+![Hash IVC Folding Scheme](attachments/hash_ivc_folding_scheme.png)
+
+### IVC Folding Scheme with zkSNARKs
+
+First, the witness $W$ and $w$ are sent by the prover, so it is not zero-knowledge. Second, the proof size is linear in the size
+of $F$. Therefore, we should use zkSNARKs to make the proof zero-knowledge and succinct.
+
+To do that, the prover creates a proof $\Pi$ claims that he knows a witness $W'$ such that:
+$(U', W') = folding((U, W), (u, w))$, and the verifier will verify that proof for each iteration.
+
+![zkSNARK IVC folding scheme](attachments/zk_snark_ivc_folding_scheme.png)
+
+Unfortunately, employing an off-the-shelf zkSNARK makes the overall solution impractical. The reason is that the zkSNARK prover
+must prove the knowledge of vectors whose commitments equal a particular value (e.g., $\bar W$ is the commitment of $W$). This
+requires encoding a linear number of group scalar multiplications in the programming model of zkSNARKs. One way to make this
+practical is by interpreting commitments to vectors as polynomial commitments.
+
+## References
+
+[Nova paper - Abhiram Kothapalli & Srinath Setty & Ioanna Tzialla](https://eprint.iacr.org/2021/370.pdf)
+
+[Nova: IVC from folding scheme - ZKstudy Session 07 [EN]](https://www.youtube.com/watch?v=xpgDdTqPnSg&t=1389s)
+
+[(Workshop) [Super] Nova [Scotia]: Unpacking Nova](https://www.youtube.com/watch?v=N6RW_YhLMNw)