Correcting any single error: Shor [[9,1,3]]
In Section 13.9 we derived the encoding circuit for the Shor [[9,1,3]]-code, so now let’s go from the top and put all the pieces together to understand how this gives an error correction procedure for all possible single-qubit errors.
To start, we encode our qubit with the phase-flip code
and then we encode each of the resulting three qubits with the bit-flip code
resulting in a net effect of
In order to understand how this code works, it is helpful to look at the complete circuit diagram, so let’s build it up in a compositional way.
Rather than drawing the entire circuits from Sections 13.7 and 13.8 again, let’s simply draw them as consisting of an encoding gate C and a decoding gate D, separated by a zone [t_1,t_2] that corresponds to transmission over the noisy channel.
Then the bit-flip correction circuit looks like
and the phase-flip correction looks the same, but with Hadamard gates sandwiching the transmission zone, so
Then we can nest the bit-flip correction circuit into the phase-flip correction circuit by inserting a copy on each of the three wires in the transmission zone, giving us the circuit that implements the encoding of Equation (\ddagger), as shown in Figure 13.11.
Figure 13.11: Nesting the two correction circuits: one copy of the bit-flip correction circuit on each wire of the phase-flip correction circuit.
Now, if an X error occurs on one of the nine qubits in the circuit in Figure 13.11 during the time interval [t_1,t_2], it will be corrected by the corresponding inner three-qubit repetition code that corrects for bit-flips.
In fact, this scheme can tolerate up to three bit-flip errors provided that they occur in different blocks.
For example, writing X_i to mean an X error on the i-th qubit, if X_1, X_5, and X_7 all occur then they will all be corrected, but if X_1 and X_2 both occur then the resulting error will not be corrected.
Next, if a Z error occurs on one of the qubits, say the first one, we know that it will not be corrected by the inner encoding–decoding circuit (the one taking place on the top three qubits), but it will be passed along and then corrected “one level up”, by the outer encoding–decoding circuit (the one taking place on the first, fourth, and seventh qubits).
Finally, what about if a Y error occurs?
Well, since Y=ZX, the inner circuit will correct the X part of the error, and the outer circuit will correct the Z part
So quantum error correction is indeed possible: we can remove the unwanted effects of decoherence during transmission through a channel.
However, this process of encode–transmit–decode doesn’t really cover the practical scenario of computation, since in reality we are constantly trying to process our data, and noise could enter at any moment.
One thus has to compute on the encoded states the whole time, whilst also somehow ensuring that even faults occurring during an error correction cycle don’t adversely affect the computation.
This is known as fault tolerance, and studying this, using the stabiliser formalisation of Section 7, is the goal of Section 14.