Text from the interactive explanation
Quantum Topological Error Correction Codes: The Classical-to-Quantum Isomorphism Perspective
Daryus Chandra, Zunaira Babar, Hung Viet Nguyen, Dimitrios Alanis, Panagiotis Botsinis, Soon Xin Ng, Lajos Hanzo
Original paperQuantum topological error-correction codes
Quantum computers will need error correction to run reliably. But the classic trick—keep copies and compare them—is off the table: quantum rules forbid copying a qubit, and looking at one destroys what it holds.
This paper shows how laying qubits on a grid, with every check touching only its neighbours, turns classical grid codes into quantum ones, and works out how much noise they can take. That locality matters: the IBM chips the paper cites (5- and 16-qubit devices) only let neighbouring qubits interact—exactly what grid codes need.
On the right, a 41-qubit surface code keeps watch. Noise strikes, red and blue lamps report only odd or even, and a decoder guesses the most likely culprit—without ever seeing the errors. Tap a bead to plant your own.
qubit
X error (bit flip)
Z error (phase flip)
Y = both
red lamp: sees Z
blue lamp: sees X
decoder fix
d = 5 · 41 qubits · corrects any 2 errors
Watching: each round, noise strikes, the lamps report, and the decoder repairs.
An unprotected qubit at this noise is hit in about 4% of rounds.
Live simulation on this page following the paper’s depolarizing-noise model; a minimum-weight decoder picks the most likely error. Illustrative, not the paper’s data.
Classical topological codes · Section II
A grid that checks itself
Start with ordinary bits. Put 13 of them on the edges of a small square net (the beads). At each crossing sits a parity lamp that watches the three or four bits touching it and lights up when an odd number of them has flipped. That is the whole code: the paper’s Fig. 4 and Table I.
The pattern of lit lamps is called the syndrome. It never tells you the stored message—only where the trouble is. Try it.
Tap a bead (or focus it and press Enter) to flip that bit. Arrow keys move between beads.
Pick a challenge, or flip any bits you like.
All lamps dark: nothing to fix.
A distance-d code can always fix t = ⌊(d−1)/2⌋ errors. This grid has d = 3, so t = 1: one flipped bit is always recovered, two can fool it. The decoder simply picks the fewest flips that explain the lamps—with a small error rate, one flip is far more likely than two.
Flipping a whole row from edge to edge lights nothing: every lamp sees two flips. That pattern is itself a legitimate codeword, so no check can ever notice it. Hold on to that picture—a chain from edge to edge—it returns in chapter 4.
Grow the grid
Bigger nets protect better: the distance d equals the lattice size. But look at what each bit gets.
Rate r = k/n settles at ½ while the protection per bit δ = d/n shrinks toward 0 (paper Eqs. 8–10, 18–20; Figs. 8–9).
Lattice codes trade per-bit strength for structure: rate ½ and unbounded distance, but δ → 0 as they grow. Hamming codes have d = 3; BCH codes shown have n = 255. Curves: Hamming bound r ≤ 1 − H(δ/2) and Gilbert–Varshamov bound r ≥ 1 − H(δ). The authors leave the reason lattice codes sit where they do as an open question.
Now swap the bits for qubits. Three things break.
Quantum errors and stabilizers · Section III
Why quantum is harder
No copying
The no-cloning theorem forbids making a copy of an unknown qubit. The oldest trick—store three copies and vote—is gone.
Looking destroys
Measuring a qubit collapses its superposition into a plain 0 or 1. You cannot peek at the data to check it.
Two kinds of error
Besides the bit flip X there is a phase flip Z, which has no classical twin—and Y is both at once.
One qubit, up close
0.97|0⟩ + 0.26|1⟩
This is the standard textbook picture of one qubit, sliced flat: up is |0⟩, down is |1⟩, and the sideways directions are equal mixtures whose sign (the phase) differs. Apply errors and watch the arrow.
X mirrors top↔bottom (swaps the amounts of |0⟩ and |1⟩). Z mirrors left↔right (flips the sign). Measuring snaps the arrow up or down with probabilities set by the amounts squared.
The trick: ask the group, not the member
Four qubits hide behind frosted glass around one lamp. Tap any of them to flip it. The lamp answers a single question—is the number of flips odd or even?—and reports +1 or −1. It never reveals which qubit flipped, nor what any qubit holds.
+1 (even) → syndrome bit 0
This is a stabilizer measurement, the quantum cousin of the parity lamp: eigenvalue +1 ↔ syndrome 0, −1 ↔ syndrome 1 (paper Eq. 44).
The translator
X → (0 | 1)
Noticed by the blue face lamps.
Every quantum error on one qubit can be written as two classical bits: a Z part and an X part (paper Eq. 48). Errors become bit patterns, and the checks become two classical parity-check matrices—one hunting X errors, one hunting Z errors.
This translation is the classical-to-quantum isomorphism : it lets classical code designs be turned into quantum ones. The catch is that the pair of check matrices must satisfy the symplectic criterion (Eq. 50) so that all the checks can be measured together without disturbing one another.
Surface codes and their family · Section IV
One lattice, two sets of lamps
Here is the same 13-bead net from chapter 1, now holding qubits. The red squares on the crossings are still there; they catch Z errors. A second set of lamps— blue triangles in the faces—catches X errors. This is the distance-3 surface code of the paper’s Fig. 12 and Table VIII.
Choose X, Z or Y, then tap a qubit (or focus it and press Enter).
All lamps dark.
The rate bill
The classical grid stored k = 7 bits in 13. Its quantum version stores just one qubit in 13: two lamp families eat the redundancy. The isomorphism gives the bill exactly: rQ = 2·rC − 1 = 2·7/13 − 1 = 1/13 (Eq. 57).
As the lattice grows, rC → ½, so rQ → 0 (Fig. 14).
The family
The same recipe works on other lattices. Pick one; numbers follow Table X at the distance chosen above.
Why grids suit real chips. On IBM’s 16-qubit ibmqx5, two-qubit interactions are possible only between neighbouring qubits. Grid codes need nothing more: every check involves only nearby qubits (paper Section I).
Layout redrawn after Fig. 1(c) of the paper (coupling directions omitted).
Degeneracy · Section V, Figs. 16–17
Be the decoder
A distance-5 surface code, 41 qubits. Somebody planted Z errors you can’t see; you see only which red lamps are lit. Tap qubits to place your fix (violet rings) until every lamp goes dark—then reveal what really happened.
Tap a qubit to toggle a Z fix on it. Arrow keys move, Enter toggles.
Three different errors light exactly these lamps. Which one happened? Does it matter?
2 lamps lit · your fix: 0 qubits
The verdict appears when you reveal the hidden error.
Quantum codes have three kinds of outcome, and one of them has no classical twin (Section V):
Lamps light. The decoder knows something is wrong and gets a chance to fix it.
A chain from edge to edge. Every lamp sees two flips, so none lights—yet the stored qubit has changed.
A closed loop is itself a stabilizer: it leaves the stored state untouched. This is the quantum bonus called degeneracy.
The paper calls these codes highly degenerate: most syndromes are ambiguous. In the distance-3 surface code, only 13 of the 64 possible red-lamp patterns correspond to a single error. In the authors’ simulations of distance-3 codes, crediting harmless loops as successes lowered the measured qubit error rate (Fig. 18)—though their simulation could only partly enumerate these loops.
Thresholds and fidelity · Sections V-A, V-C
Turn up the noise
The paper measures success by the qubit error rate (QBER): the probability that the protected qubit ends up wrong. The curves below come from its worst-case upper bound (Eq. 60), which counts every pattern of more than t errors as a failure:
QBERupper(n, d, p) = 1 − Σi=0..t C(n,i) pi(1−p)n−i − S·p4(1−p)n−4, t = ⌊(d−1)/2⌋, S = number of stabilizers
So real performance is at least this good; a live decoder like the one at the top of the page also rescues some heavier patterns. The paper approximates all stabilizer weights as 4.
The dashed line marks the paper’s reported threshold for the chosen family: 1.83×10⁻² (colour), 1.34×10⁻² (rotated surface), 6.28×10⁻³ (surface), 6.77×10⁻³ (toric). Pairwise crossings of the curves drift slightly with d, so think of it as a crossover zone rather than a single knife-edge.
The fidelity bar
Flip the question: how good must the hardware already be before a code helps at all? With initial fidelity Fin = 1 − p, the code’s output fidelity is Fout = 1 − QBERupper. Above the diagonal, the code improves things; below it, the code makes them worse.
Break-even points from the paper (Fig. 22): 0.942 for the colour code (7 qubits), 0.968 for the rotated surface code (9), 0.986 for the surface code (13) and 0.993 for the toric code (18). The hardware must already be good before a code helps—and fewer physical qubits lower that bar.
The hashing bound · Section V-B
The price of protection
Bigger lattices lower the error rate—but they spend more qubits to store the same single qubit. To compare fairly, the paper measures each code against the quantum hashing bound , CQ(p) = 1 − H(p) − p·log₂3 (Eq. 61): the most noise any code of rate rQ could in principle survive, p*.
The distance from the hashing bound, D = p* − p (Eq. 62), says how far below that limit a code needs the noise to be. Ideal codes work near D = 0. The worst possible is D0 = 0.1893, where the bound itself hits zero rate (Eq. 63).
Growing the lattice improves QBER against p, but the extra qubits cost more than the distance gains: the curves crowd toward D0 instead of marching toward D = 0, and the rate rQ heads to zero. The paper’s conclusion: these codes are best suited to short-to-moderate lengths. Its suggested routes closer to the bound are better constructions at the same rate, or more powerful decoders such as soft-decision ones.
Conclusions
Back to the opening question
How do you fix a mistake you’re not allowed to look at? You ask groups of neighbours only odd or even—and let the pattern of answers point to the most likely culprit.
Parity lamps. Bits on a lattice, checks at the crossings: the syndrome locates errors without reading the message.
The isomorphism. Each quantum error becomes two classical bits, so a pair of classical check matrices becomes a quantum code.
A family that needs only neighbours. Colour, rotated-surface, surface and toric codes—matched to chips where only neighbours interact.
Degeneracy. Different errors can be equally good to fix; closed loops are harmless, edge-to-edge chains are not.
Thresholds and the fidelity bar. Below a noise threshold bigger lattices win; but hardware must already be good, and growing the lattice costs more qubits than it gains in distance.
A quick test for any code
For hard-decision, syndrome-decoded stabilizer codes, the paper distils its bound into a one-line estimate of the break-even fidelity (Eq. 70): Fth(n, d) ≈ 1 − C(n, t+1)−1/t, with t = ⌊(d−1)/2⌋.
An upper-bound approximation: for the colour code it gives 0.952, against 0.942 from the full bound. Try n = 25, d = 5 (rotated surface).
Source & credits
Daryus Chandra, Zunaira Babar, Hung Viet Nguyen, Dimitrios Alanis, Panagiotis Botsinis, Soon Xin Ng, Lajos Hanzo, “Quantum Topological Error Correction Codes: The Classical-to-Quantum Isomorphism Perspective.” Author manuscript hosted by the University of Southampton. Published in IEEE Access, vol. 6, pp. 13729–13757 (2018): doi:10.1109/ACCESS.2017.2784417.
All diagrams are redrawn with code from the paper’s descriptions (Figs. 1, 4, 5, 12, 13, 16, 17) and all curves are computed from the paper’s Eqs. 10, 16, 18, 57, 60–63 and 70. The opening lattice and the decoder game are live simulations built for this page using the paper’s noise model; they illustrate the mechanism and are not the paper’s data.
The thresholds, break-even fidelities and degeneracy results quoted here come from the paper’s simulations and analytical bounds, not from hardware experiments.
Topics in this iPaper
Physics
One of the essential prerequisites to build quantum computers is the employment of quantum error correction codes (QECCs)
Section I Introduction, page 1
Computer science
The formulation of QSCs is capable of detecting both the number and the position of errors without actually observing the state of physical qubits
Section III-C, page 13
Engineering
We conceive the construction of classical error correction codes based on topological or lattice structures.
Section II, page 3
quantum error correction
we char-acterize the performance of QTECCs in the face of the quantum depolarizing channel in terms of both the quantum-bit error rate (QBER) and fidelity
Abstract, page 1
topological codes
the QTECCs, constitute a member of the QSC family, whose stabilizer operators are defined by the underlying lattice structure
Section IV, page 14
surface codes
Explicitly, surface codes represent the quantum equivalent of classical TECCs on rectangular lattice structures.
Section IV, page 14
stabilizer codes
The QSCs rely on the stabilizer operators Si for identifying the type, the number and also the position of the qubit errors.
Section III-C, page 13
low density parity check codes
Hence, classical TECCs can be viewed as a specific family of LDPC codes.
Section II, page 7