This tutorial will guide you in creating your first zero-knowledge SNARK circuit. It will take you through the various techniques to write circuits and show you how to create and verify proofs off-chain and on-chain on Ethereum.
If you don't have it installed yet, you need to install Node.js
.
You should install at least version 10 of node. It's important to note here that the latests versions of javascript, includes big integer support and web assembly compilers that make the code run fast.
Run:
npm install -g circom
npm install -g snarkjs
Let's create a circuit that will allow you to prove that you are able to factor a number!
- Create an empty directory called
factor
where you will put all the files that you will use in this tutorial.
mkdir factor
cd factor
In a real circuit, you will probably want to create a
git
repository with acircuits
directory and atest
directory with all your tests, and the needed scripts to build all the circuits.
- Create a new file named
circuit.circom
with the following content:
template Multiplier() {
signal private input a;
signal private input b;
signal output c;
c <== a*b;
}
component main = Multiplier();
This circuit has 2 private input signals named a
and b
and one output named c
.
The only thing that the circuit does is forcing the signal c
to be the value of a*b
After declaring the Multiplier
template, we instantiate it with a component namedmain
.
Note: When compiling a circuit, a component named main
must always exist.
We are now ready to compile the circuit. Run the following command:
circom circuit.circom --r1cs --wasm --sym
The --r1cs
option will generate circuit.r1cs
(the r1cs constraint system of the circuit in binary format).
The --wasm
option will generate circuit.wasm
(the wasm code to generate the witness).
The --sym
option will generate circuit.sym
(a symbols file required for debugging or if you want to print the constraint system in an annotated mode).
Now that the circuit is compiled, we will continue with snarkjs
.
Please note that you can always access the help of snarkjs
by typing:
snarkjs --help
To show general statistics of this circuit, you can run:
snarkjs r1cs info circuit.r1cs
You can also print the constraints of the circuit by running:
snarkjs r1cs print circuit.r1cs circuit.sym
Setup must be a trusted setup ceremony in snarks.
Please visit https://github.com/iden3/snarkjs to see how to run a ceremony for a real circuit.
To simplify it, we will run the ceremony ourself.
First we download a power of tau ceremony file:
wget https://hermez.s3-eu-west-1.amazonaws.com/powersOfTau28_hez_final_10.ptau
Then we create the zkey file withou any contribution
snarkjs zkey new circuit.r1cs powersOfTau28_hez_final_10.ptau circuit_0000.zkey
We now add out contribution
snarkjs zkey contribute circuit_0000.zkey circuit_final.zkey
Finally, we can export the verification key from the zkey file
snarkjs zkey export verificationkey circuit_final.zkey verification_key.json
Before creating any proof, we need to calculate all the signals of the circuit that match (all) the constraints of the circuit.
circom
generates a wasm module that calculates those for you. You need to provide a file with the inputs and it will execute the circuit and calculate all the intermediate signals and the output. This set of signals is the witness.
The zero-knowledge proofs prove that you know a set of signals (witness) that match all the constraints without revealing any of the signals except the public inputs and the outputs.
For example, imagine you want to prove you are able to factor the number 33. It means that you know two numbers a
and b
that when you multiply them, it results in 33.
Of course you can always use the number one and the same number as
a
orb
. We will deal with this problem later.
So you want to prove that you know 3 and 11.
Let's create a file named input.json
{"a": 3, "b": 11}
Now let's calculate the witness:
snarkjs wtns calculate circuit.wasm input.json witness.wtns
To see the witness.wtns file, you can export it to jeson and take a look
snarkjs wtns export json witness.wtns witness.json
cat witness.json
If the circuit has any error, you can debug the generation of the witness with
snarkjs wtns debug circuit.wasm input.json witness.wtns circuit.sym
Now that we have the witness generated, we can create the proof.
snarkjs groth16 prove circuit_final.zkey witness.wtns proof.json public.json
This command will use the circuit_final.zkey
and the witness.wtns
files by default to generate proof.json
and public.json
The proof.json
file will contain the actual proof and the public.json
file will contain just the values of the public inputs and the outputs.
To verify the proof run:
snarkjs groth16 verify verification_key.json public.json proof.json
This command will use verification_key.json
, proof.json
and public.json
to verify that is valid.
Here we are verifying that we know a witness that the public inputs and the outputs matches the ones in the public.json
file.
If the proof is ok, you will see OK
or INVALID
if not ok.
snarkjs zkey export solidityverifier circuit_final.zkey verifier.sol
This command will take the circuit_final.zkey
and generate solidity code in verifier.sol
file.
You can take the code in verifier.sol
and cut and paste it in remix.
This code contains two contracts: Pairings and Verifier. You only need to deploy the Verifier
contract.
You may want to use a test net like Rinkeby, Kovan or Ropsten. You can also use the Javascript VM, but in some browsers the verification takes long and it may hang the page.
The verifier contract deployed in the last step has a view
function called verifyProof
.
This function will return true if the proof and the inputs are valid.
To facilitate the call, you can use snarkjs
to generate the parameters of the call by typing:
snarkjs zkey export soliditycalldata public.json proof.json
Just cut and paste the output to the parameters field of the verifyProof
method in Remix.
If every thing works ok, this method should return true.
If you change any bit in the parameters, the result will be verifiably false.
We can fix the circuit to not accept the number 1 as any of the input values by adding some extra constraints.
Here, the trick is that we use the property that 0 has no inverse. So (a-1)
should not have an inverse.
That means that (a-1)*inv = 1
will be inpossible to match if a
is 1.
We just calculate inv by 1/(a-1)
.
So, let's modify the circuit:
template Multiplier() {
signal private input a;
signal private input b;
signal output c;
signal inva;
signal invb;
inva <-- 1/(a-1);
(a-1)*inva === 1;
invb <-- 1/(b-1);
(b-1)*invb === 1;
c <== a*b;
}
component main = Multiplier();
A nice thing of the circom language is that you can split a <==
into two independent actions: <--
and ===
.
The <--
and -->
operators assign a value to a signal without creating any constraints.
The ===
operator adds a constraint without assigning any value to a signal.
The circuit also has another problem: the operation works in Z_r
, so we need to guarantee the multiplication does not overflow. This can be done by converting the inputs to binary and checking the ranges, but we will reserve it for future tutorials.
Another problem of the circuit is that circom works with a field of a prime that in general is arround the 255bits. That means that it's very easy to factor in that field.
One possible solution to this, is to limit the inputs to 64 bits. that means that this way it will not be possible to have overflow.
The final circuit would look like:
template CheckBits(n) {
signal input in;
signal bits[n];
var lc1=0;
var e2=1;
for (var i = 0; i<n; i++) {
bits[i] <-- (in >> i) & 1;
bits[i] * (bits[i] -1 ) === 0;
lc1 += bits[i] * e2;
e2 = e2+e2;
}
lc1 === in;
}
template Multiplier(n) {
signal private input a;
signal private input b;
signal output c;
signal inva;
signal invb;
component chackA = CheckBits(n);
component chackB = CheckBits(n);
chackA.in <== a;
chackB.in <== b;
inva <-- 1/(a-1);
(a-1)*inva === 1;
invb <-- 1/(b-1);
(b-1)*invb === 1;
c <== a*b;
}
component main = Multiplier(64);
You may want to read the README to learn more features about circom
.
You can also check a library with many basic circuits lib binarizations, comparators, eddsa, hashes, merkle trees etc here (Work in progress).
Or a exponentiation in the Baby Jubjub curve here (Work in progress).
There is nothing worse for a dev than working with a buggy compiler. This is a very early stage of the compiler, so there are many bugs and lots of work needs to be done. Please have it present if you are doing anything serious with it.
And please contact us for any isue you have. In general, a github issue with a small piece of code with the bug is very useful to us.
Enjoy zero-knowledge proving!