In this post, we recall the notion of a polynomial and explain the concept of "blind evaluation" of a polynomial, and how it is implemented using Homomorphic Hiding (HH). (See part 1 for an explanation of HH). In future posts, we will see that blind evaluation is a central tool in SNARK constructions.

We denote by $\mathbb{F}_p$ the field of size $p$; that is, the element of $\mathbb{F}_p$ are {0,..., $p$ - 1}, and addition and multiplication are done mod $p$, as explained in part 1.

Polynomials and Linear Combinations

Recall that a polynomial $P$ of degree $d$ over $\mathbb{F}_p$ is an expression of the form for some $a_0,\ldots, a_d \in \mathbb{F}_p$.

$P(X) =$ $a_0 + a_1 \cdot X +$ $a_2 \cdot X^2 +$ $\ldots +$ $a_d \cdot X^d$

We can evaluate $P$ at a point $s \in \mathbb{F}_p$ by substituting $s$ for $X$ and computing the resultant sum

$P(s) =$ $a_0 + a_1 \cdot s +$ $a_2 \cdot s^2 +$ $\ldots + a_d \cdot s^d$

For someone who knows $P$, the value $P(s)$ is a linear combination of the values $1, s,\ldots,s^d$, where linear combination just means "weighted sum"; in the case of $P(s)$, the "weights" are $a_0,\ldots,a_d$.

In the last post, we saw the HH $E$ defined by $E(x) = g^x$, where $g$ was a generator of a group with a hard discrete log problem. We mentioned that this HH "supports addition" in the sense that $E(x + y)$ can be computed from $E(x)$ and $E(y)$. We note here that it also "supports linear combinations", meaning that, given $a, b, E(x), E(y)$, we can compute $E(ax + by)$. This is simply because

$E(ax+by) =$ $g^{ax+by} =$ $g^{ax} \cdot g^{by} =$ $(g^x)^a \cdot (g^y)^b =$ $E(x)^a \cdot E(y)^b$

Blind Evaluation of a Polynomial

Suppose Alice has a polynomial $P$ of degree $d$, and Bob has a point $s \in \mathbb{F}_p$ that he chose randomly. Bob wishes to learn $E(P(s))$,i.e., the HH of the evaluation of $P$ at $s$. Two simple ways to do this are:

• Alice sends $P$ to Bob, and he computes $E(P(s))$ by himself.
• Bob sends $s$ to Alice; she computes $E(P(s))$ and sends it to Bob.

However, in the blind evaluation problem, we want Bob to learn $E(P(s))$ without learning $P$, which precludes the first option; and, most importantly, we don't want Alice to learn $s$, which rules out the second[1]

Using HH, we can perform blind evaluation as follows:

• Bob sends to Alice the hidings $E(1)$, $E(s)$,...,$E(s^d)$.
• Alice computes $E(P(s))$ from the elements sent in the first step, and sends $E(P(s))$ to Bob. Alice can do this since $E$ supports linear combinations, and $P(s)$ is a linear combination of 1, $s$,...,$s^d$.

Note that, as only hidings were sent, neither Alice learned $s$ [2], nor did Bob learn $P$.

Why Is This Useful?

Subsequent posts will go into more detail on how blind evaluation is used in SNARKs. The rough intuition is that the verifier has a "correct" polynomial in mind and wishes to check that the prover knows it. Making the prover blindly evaluate their polynomial at a random point not known to them ensures the prover will give the wrong answer with high probability if their polynomial is not the correct one. This, in turn, relies on the Schwartz-Zippel Lemma, which states that "different polynomials are different at most points".