Quantum Repeaters

Alice in Bengaluru wants a Bell pair with Bob in Delhi. The straight-line distance is about 1750 km; through a real fibre route, more like 2200 km. Each kilometre of silica fibre absorbs about 0.2 decibels of a 1550 nm photon — meaning roughly 4.5% loss per kilometre, compounded. After 100 km, about 1% of the photons survive. After 500 km, about one in 10^{10}. After 1000 km — let alone 2200 km — the survival rate is so small that a photon source running at gigahertz would deliver a single photon to Delhi maybe once per month, if ever.

The classical internet does not have this problem. When a fibre-optic signal weakens, a repeater station every 50 km reads the signal, amplifies it, and re-transmits. A quantum signal cannot be repaired this way: measuring the photon at the halfway point would collapse the quantum state and destroy the qubit; copying the photon without measurement is forbidden by the no-cloning theorem (ch.20). So the whole classical playbook is out.

What replaces it is the idea you derived in the entanglement-swapping chapter: you can stitch short Bell pairs into long ones without any qubit traversing the full distance. A quantum repeater is just that idea, industrialised: a chain of nodes, each of which holds short-hop Bell pairs in quantum memory, performs Bell measurements to swap entanglement across itself, and — when the pairs are noisy enough to worry about — runs an entanglement-distillation protocol to clean them up before swapping. This chapter derives why each of those pieces is necessary, walks through the DLCZ architecture that made them concrete, counts the rate budget for a chain, and shows where the ISRO Bengaluru–Mt. Abu satellite-QKD demonstration (2022) and China's Micius satellite fit in the global roadmap.

The problem — exponential loss in fibre

Picture first. Here is what happens to a stream of photons pushed into a long optical fibre.

The physics is clean. Attenuation in fibre is a per-length probability of absorbing or scattering each photon. A single photon has survival probability

where \alpha = 0.2 dB/km and L is the length in km.

Why the 10^{-\alpha L/10} form: the decibel is a logarithmic unit. X dB of loss means the intensity ratio is 10^{X/10}. For 0.2 dB/km over L km, total loss in dB is 0.2L, and the surviving fraction is 10^{-0.2L/10}.

Plugging in numbers: at L = 50 km, survival is 10^{-1} = 0.10 (10% make it). At L = 100 km, 10^{-2} = 0.01. At L = 500 km, 10^{-10}. At L = 1000 km, 10^{-20}.

Now suppose you have a single-photon source that fires at R_0 = 10^9 photons per second (a fast, lab-quality pulsed source). At L = 500 km, the rate at the far end is

One photon every 10 seconds. At L = 1000 km the rate is 10^{-11} per second — roughly one photon per three thousand years. Direct transmission is dead past about 500 km.

A trivial "fix" — amplify the signal at a halfway station — runs straight into the no-cloning theorem. The classical amplifier is modelled as a map |\psi\rangle \to |\psi\rangle \otimes |\psi\rangle (turn one photon into two identical ones). That map is not unitary for unknown |\psi\rangle: if it worked for both |0\rangle and |+\rangle, linearity would force it to take |0\rangle + |+\rangle to |0\rangle\otimes|0\rangle + |+\rangle\otimes|+\rangle, which does not equal (|0\rangle+|+\rangle)\otimes(|0\rangle+|+\rangle). The only way to amplify a photon is to measure it, which collapses the quantum state. So classical repeater logic fails.

The repeater idea — short segments, swap at the nodes

Break the long distance into N short segments of length L_0 = L/N. On each segment, share a Bell pair — that photon only has to survive one hop of length L_0, so the survival probability is manageable. Then at every intermediate node, perform entanglement swapping (ch.52) on the two Bell pairs that meet there, stitching them into one longer pair. After all N-1 swaps, Alice and Bob share a single Bell pair spanning the full distance, with no qubit having traversed more than L_0 km.

The naive rate calculation already shows why this is a win over direct transmission. If each segment has photon survival probability p_0 = 10^{-\alpha L_0 / 10}, then the probability that all N segments succeed simultaneously (ignoring memories, for now) is p_0^N. For L = 500 km split into N = 10 segments of L_0 = 50 km:

Which is the same as the direct-transmission rate — no gain yet. The gain only shows up when you add quantum memories: each segment can try, fail, retry independently; store its Bell pair as soon as the segment succeeds; wait for the other segments; then do the swaps. The rate then scales polynomially with N (roughly as 1/N^2 for a simple memory-assisted protocol, as opposed to exponentially with L).

Why memories are the magic ingredient: without memory, all segments must succeed simultaneously, and the joint probability is the product — exponentially small. With memory, each segment succeeds at its own pace, the expected time is set by the slowest segment, and failures on one segment don't waste the other segments' successes.

This polynomial-versus-exponential gap is the whole payoff. A repeater chain with good memories can reach continental distances; direct fibre cannot.

Entanglement distillation — cleaning up before swapping

There is a second problem, more subtle than loss. Real Bell pairs are noisy. A photon-pair source might produce pairs of fidelity F = 0.98 — meaning the state is a 0.98 mixture of the ideal |\Phi^+\rangle with a 0.02 admixture of other Bell states or junk. Each swap compounds the noise: if two input pairs have fidelity F, the swapped pair has fidelity roughly F^2 (ignoring exact noise models). After a chain of N swaps, fidelity drops like F^N, and by the time F^N falls below 0.5 or so, the "Bell pair" is no better than classical correlation.

Entanglement distillation (ch.entanglement-as-resource) fixes this. You take k noisy Bell pairs, perform a local protocol (a CNOT between the pairs and a measurement on one of them), and output m < k cleaner pairs of higher fidelity. Distillation is a probabilistic upgrade: inputs with fidelity F = 0.8 might yield one output pair of F' = 0.95 for every two input pairs consumed, succeeding half the time.

A realistic repeater protocol therefore alternates three moves at each node: attempt segment (probabilistic success, heralded by a detector click), distill (combine several successful segment-pairs into fewer higher-fidelity pairs), swap (Bell-measure to extend the pair). The architecture choices — how much you distill, when, with what protocol — are the main axis along which different repeater proposals differ.

Three generations of architectures are conventionally distinguished. First-generation repeaters use probabilistic heralded entanglement generation and entanglement distillation (much slower but tolerant of loss and gate errors — DLCZ is the archetype). Second-generation repeaters add quantum error correction on small codes to handle operation errors more efficiently than distillation. Third-generation ("all-photonic") repeaters encode each logical qubit in a large cluster state so that loss events within the cluster can be tolerated without waiting for retries — potentially removing the need for long-lived memories. All three are active research directions in 2026, with first-generation systems the only ones at lab demonstration and early deployment.

The DLCZ protocol — atomic ensembles and heralded entanglement

The landmark paper on practical repeaters is Duan, Lukin, Cirac, and Zoller, Long-distance quantum communication with atomic ensembles and linear optics, Nature 414, 413 (2001), arXiv:quant-ph/0105105. It introduced a concrete architecture that (i) uses atomic ensembles as memories, (ii) generates entanglement through a heralded single-photon process so that loss events are detected rather than silently destroying the protocol, and (iii) chains swaps and distillations in a well-defined sequence. Most modern first-generation repeater proposals are DLCZ variants.

The DLCZ segment works like this. Each of the two nodes on the segment has an atomic ensemble — a cloud of about 10^{10} cold atoms in a magneto-optical trap. A weak write-laser pulse drives the ensemble so that with small probability p \ll 1 a single collective excitation is created in the ensemble, and a single photon is emitted in a well-defined mode. Both nodes pulse simultaneously. Each emitted photon travels down its fibre to a 50:50 beam splitter placed halfway between the two nodes. On the far side of the beam splitter sit two single-photon detectors.

Here is the clever bit. If exactly one of the two detectors clicks, the protocol heralds that a single collective excitation was created — and because the beam splitter has erased which-path information (the photon could have come from either ensemble), the two ensembles are now entangled: one has the collective excitation, the other is in its ground state, with a superposition over which one has the excitation. The Bell-pair state lives in the two ensembles, stored in the long-lived atomic coherence. If zero or two detectors click, the protocol failed, discard and retry.

The key win here is heralding: the photon's trip through the fibre happens before you know whether you have a Bell pair. If the photon is lost in the fibre, neither detector clicks and the protocol simply discards the attempt. Loss reduces the rate of successful heralds, but it does not corrupt the Bell pairs that did succeed. So you can run the segment at whatever rate physics allows, store the heralded pairs in atomic memories, and move on.

Once both halves of a repeater chain have heralded Bell pairs on both their segments, the middle node performs an entanglement swap, again heralded by single-photon detection. The overall rate scales (for the idealised DLCZ protocol) as roughly

where p_0 is the single-ensemble emission probability, \eta_d is the detector efficiency, N is the number of segments, and T_0 is the classical communication time per segment. The explicit derivation of this rate (the Sangouard-Simon-de Riedmatten-Gisin 2011 review is the canonical source) is beyond our scope, but the key feature is that R falls polynomially with N, not exponentially. For continental distances this is the difference between "impossible" and "slow but tractable."

Satellite QKD — free space sidesteps fibre altogether

Fibre loss at 0.2 dB/km is a law of silica chemistry. Free-space loss — sending the photon through vacuum — is much gentler. Atmospheric attenuation is significant only in the lowest 10 km; above that, the photon travels essentially unimpeded. A satellite at 500 km altitude, firing a photon directly down to a ground station, has total loss of about 10–20 dB in good weather: much better than 100 dB for the equivalent fibre distance.

This is the logic behind satellite-based QKD. A satellite in low Earth orbit carries an entangled-photon source. It distributes Bell pairs by beaming one photon of each pair to one ground station and the other photon to a second ground station (or it distributes single photons for BB84). Because each ground station is only seeing the satellite for a few minutes per pass, the data rate is modest — but the per-photon loss is radically better than fibre, and critically, loss does not grow with the ground-station-to-ground-station distance (only with the slant range from each station to the satellite).

The flagship demonstration is China's Micius satellite (Mozi in Chinese), launched August 2016. Between 2016 and 2018, Micius demonstrated BB84 QKD between ground stations 1200 km apart, and entanglement distribution between pairs of ground stations with separations up to 1200 km. The key distribution was not repeater-based in the full sense — there was no entanglement swap at a ground relay — but it was the first demonstration that satellite links could beat fibre for continental-scale quantum communication.

India's corresponding effort is led by ISRO's Space Applications Centre in Ahmedabad and the Raman Research Institute in Bangalore. In 2022, ISRO demonstrated free-space quantum-key distribution over a 300-metre rooftop-to-rooftop link and subsequently over a longer atmospheric link, with the Bengaluru–Mt. Abu demonstration serving as the first Indian intercity free-space QKD test. The Mt. Abu observatory in Rajasthan is about 1100 km from Bangalore by road, though the actual 2022 experiment used a satellite relay to bridge that gap. The effort is part of the National Quantum Mission (2023, ₹6003 crore over 8 years), which has "long-distance secure quantum communication" as one of its four explicit pillars.

Satellites are not a universal substitute for a ground repeater network. You need line-of-sight to a satellite, which means clear weather and limited overhead-pass windows. You need precise beam steering (a microradian of pointing error translates to metres on the ground after 500 km). And the total data rate per pass is kilobits per second at best — enough for key distribution, not enough for high-throughput quantum communication. But for a first rung on the quantum-internet ladder, satellites let you leap over the fibre-loss wall today, while ground repeaters are still being engineered.

Common confusions

• "Quantum repeaters amplify the quantum signal." No. They do not touch the quantum signal at all — they entangle short-hop pairs and then use local operations and classical communication to stitch those pairs into a long pair. The original photon carrying the qubit is consumed by the segment Bell measurement long before anyone asks what its state is. The word "repeater" is borrowed from classical optics but is misleading; "entanglement-swapping relay" is more accurate.

• "A quantum repeater is just a classical fibre amplifier with extra steps." No. A classical amplifier boosts the signal intensity. A quantum repeater establishes entanglement between two distant parties by chaining short entangled links, using a process (Bell measurement) that has no classical analogue. The two devices share a name and a spacing strategy, but the physics is entirely different.

• "Quantum memories are a nice-to-have." They are load-bearing. Without memories, all segments of the chain must succeed simultaneously, and the probability of that is exponentially small in N (the exact scaling that you were trying to escape in the first place). With memories, each segment's successes accumulate independently, and the rate becomes polynomial in N. The whole repeater concept collapses without long-lived memories — which is why memory engineering is the critical-path bottleneck in every repeater programme in 2026.

• "Trusted-node QKD is a kind of quantum repeater." Not in the full sense. In a trusted-node network (like China's Beijing–Shanghai backbone, or the early Indian metro deployments), QKD is done pairwise between each adjacent node, and the final end-to-end key is assembled by classically relaying intermediate keys through the nodes. This requires trusting every intermediate node — if an intermediate node is compromised, the key is compromised. A true quantum repeater distributes entanglement end-to-end, so no intermediate node ever learns the key. Trusted-node networks are a practical stopgap; full quantum repeaters are the long-term goal.

• "Satellite QKD removes the need for ground repeaters." Partially. Satellites give continental-scale key distribution but only in brief overhead windows, with modest throughput. Ground repeaters aim at continuous service with much higher key rates. The mature quantum internet will likely use both: satellites for intercontinental links (where fibre is impractical) and ground repeaters for metro and intercity coverage (where continuous service matters). The 2026 deployments are a mix.

• "Once I have a long Bell pair, I have a quantum internet." A Bell pair is one primitive. The quantum internet needs routing, addressing, error handling, authentication, shared clocks, and all the other mundane infrastructure that the classical internet needed. The Wehner-Elkouss-Hanson (2018) roadmap explicitly lays out six stages from trusted QKD to fully fault-tolerant distributed quantum computing; Bell-pair distribution puts you at stage 3 out of 6.

Going deeper

The Briegel-Dür-Cirac-Zoller (BDCZ) rate bound

The foundational theoretical paper on quantum repeaters is Briegel, Dür, Cirac, Zoller, Quantum repeaters: The role of imperfect local operations in quantum communication, Phys. Rev. Lett. 81, 5932 (1998), arXiv:quant-ph/9803056. Their key result: with nested entanglement purification and entanglement swapping, the resources required to distribute a Bell pair over distance L scale as (\text{polynomial in } L) \cdot \exp(\text{constant} \cdot L_0), where L_0 is the segment length. Crucially, once L_0 is fixed (by local fidelity thresholds), the overall resource cost is only polynomial in L — no longer exponential.

The protocol is nested. Start with 2^n elementary segments each of length L_0. First, purify all segments in parallel to raise fidelity above threshold. Then swap adjacent pairs to get 2^{n-1} segments of length 2 L_0, purify those, swap again, and so on for n levels. Each purification step costs a constant factor in extra pairs; each swap is deterministic; total resource cost is polynomial in n = \log_2(L/L_0), which is polynomial in L.

The BDCZ paper was the first to show that quantum communication over arbitrary distances is possible in principle with imperfect local operations — a non-trivial result, since every gate has errors and every memory decoheres. Without it, it was not obvious that chaining swaps and distillations would converge rather than blow up.

DLCZ mathematical protocol in detail

The DLCZ paper writes each atomic ensemble as a collective excitation operator S^\dagger = (1/\sqrt{N_a}) \sum_i |s\rangle_i \langle g|, where |g\rangle and |s\rangle are the ground and stored states of each atom and N_a is the number of atoms in the ensemble. A weak write pulse produces the state

where a^\dagger creates the emitted photon. For small p, this is essentially a pair of entangled quanta: the collective atomic excitation is correlated with the emission of one photon.

On both ensembles simultaneously, the joint state is

The two emitted photons a_A^\dagger and a_B^\dagger propagate down their fibres to a 50:50 beam splitter halfway. The beam splitter transforms the modes as

After the beam splitter and a single-click event at detector D_1 (which projects onto a_A^\dagger + a_B^\dagger), the heralded atomic state is

Why the single-click heralding projects to this form: a click at D_1 means one photon arrived in the (a_A + a_B)/\sqrt 2 mode. Running the beam-splitter equations backwards, the incoming photon had to have come from a superposition of the two ensembles' emitted modes — and by momentum/mode conservation, the ensemble that "sent" the photon is the one with the atomic excitation. So the post-herald atomic state is the W-like superposition above.

That is a Bell pair: one collective excitation shared coherently between the two ensembles. Stored in atomic coherence, it can live for hundreds of microseconds to milliseconds depending on the ensemble. A follow-up read pulse converts the atomic excitation back to a photon on demand, enabling the swap operation at the next level.

The full DLCZ protocol extends this to the N-segment chain. The rate calculation, accounting for heralding probability p, detector efficiency \eta_d, and classical communication time T_0 = L_0/c, gives (to leading order)

where f(N) is a polynomial overhead that depends on how the swaps are scheduled. The Sangouard-Simon-de Riedmatten-Gisin 2011 review (Rev. Mod. Phys. 83, 33) gives the full formulas for a range of DLCZ variants.

First-, second-, and third-generation repeaters in comparison

First-generation (DLCZ-like). Probabilistic heralded entanglement generation over fibre. Entanglement distillation to boost fidelity. Bell-measurement swaps at intermediate nodes. Requires long-lived matter memories (atomic ensembles, rare-earth-doped crystals, NV centres, trapped ions). Pros: tolerant of loss, simple logic. Cons: very slow, bounded by distillation overhead and memory lifetime.

Second-generation. Replace entanglement distillation with quantum error correction on small codes. Each "Bell pair" is really a pair of logical qubits, each encoded across several physical qubits. Errors on individual physical qubits are detected and corrected via syndrome measurements; the logical Bell pair remains high-fidelity. Pros: much faster than distillation once hardware thresholds are met. Cons: gate-error thresholds are stringent; requires near-fault-tolerant local operations.

Third-generation ("all-photonic"). Encode each link's entanglement in a large cluster state (a graph of entangled photons) generated locally. Loss within the cluster is tolerated by graph-theoretic redundancy; no memory is needed between segments. Pros: potentially much faster; no long-lived memories. Cons: requires massive single-photon-source throughput; cluster-state generation at useful scale is not yet demonstrated.

As of 2026, the deployed systems are first-generation (metro-scale fibre with trusted relays, plus satellite links). Second-generation systems are in early laboratory demonstration, and third-generation is at the stage of detailed architectural studies.

India's National Quantum Mission and the networking pillar

The National Quantum Mission, announced in 2023 with a budget of ₹6003 crore over eight years, defines four verticals: computing, communication, sensing, and materials/devices. The communication pillar is the one relevant to this chapter, and its targets are:

• Metro-scale fibre QKD networks in the four major Indian metros (Delhi, Mumbai, Bengaluru, Hyderabad) by 2028.
• Intercity trusted-node QKD over fibre, connecting at least two metros by 2030.
• Satellite-relayed QKD between Indian ground stations, extending the 2022 ISRO Bengaluru–Mt. Abu demonstration, targeting full operational service by 2032.
• Long-term: true entanglement-based quantum-repeater networks replacing trusted nodes, on a 10–15 year horizon.

Participating institutions include ISRO Space Applications Centre (Ahmedabad), Raman Research Institute (Bengaluru), TIFR (Mumbai), IIT Madras, IISc Bangalore, IIT Bombay, and IIT Delhi, plus industrial partners (HCL Tech, TCS, QNu Labs). The architectural plan is essentially a first-generation repeater network with satellite overlays, following the template established by Micius and the Beijing-Shanghai backbone, adapted for Indian geography and the specific challenges of monsoon-season free-space links.

The engineering stakes are concrete. Aadhaar authentication and UPI payments rely on cryptographic signatures based on RSA and ECC — public-key systems that Shor's algorithm (when a large enough quantum computer exists) will break. Post-quantum cryptography is the main migration path, but QKD-backed keys are an additional layer, particularly for the most sensitive inter-agency communications. The NQM's communication pillar is in that sense a strategic cryptographic hedge, not merely a science project.

Why repeaters are harder than they look

It is worth naming the specific engineering obstacles. (i) Quantum memory lifetimes are currently in the millisecond range for atomic ensembles and hundreds of milliseconds for rare-earth crystals — adequate for lab-scale but not for intercontinental timing. (ii) Photon-to-memory coupling efficiency ("write" and "read") is typically 30–50%; every inefficiency multiplies into the rate budget. (iii) Linear-optical Bell measurements (Lütkenhaus-Calsamiglia, 1999) are at most 50% efficient without ancillary entanglement; using hybrid platforms (photon for communication, matter qubit for the Bell measurement) is essential. (iv) Wavelength conversion — most memories operate at alkali-atom wavelengths (780 or 852 nm), while telecom fibre is at 1550 nm — requires nonlinear-optical frequency converters that are not yet standard lab equipment. (v) Synchronisation across N nodes, each with its own clock, is non-trivial when the protocol demands picosecond-precision timing of Bell-measurement events. All five are active research problems; none is solved. The 10–15 year NQM target for a full quantum-repeater network is a realistic engineering horizon, not an optimistic guess.

Where this leads next

• The Quantum Internet — the vision that repeater chains stitch into a global network.
• Entanglement Swapping — the primitive that every repeater node performs.
• Entanglement as a Resource — including distillation, the other primitive repeaters depend on.
• A Tiny Quantum Network — the three-node minimal case of a repeater chain.
• BB84 Protocol — the QKD application that motivates most repeater deployments.
• No-Cloning Theorem — the theorem that forbids the classical-amplifier shortcut.

References

1. L.-M. Duan, M. D. Lukin, J. I. Cirac, P. Zoller, Long-distance quantum communication with atomic ensembles and linear optics (2001) — arXiv:quant-ph/0105105. The DLCZ paper.
2. H.-J. Briegel, W. Dür, J. I. Cirac, P. Zoller, Quantum repeaters: The role of imperfect local operations in quantum communication (1998) — arXiv:quant-ph/9803056. The foundational BDCZ paper showing polynomial resource scaling is possible.
3. N. Sangouard, C. Simon, H. de Riedmatten, N. Gisin, Quantum repeaters based on atomic ensembles and linear optics (2011) — arXiv:0906.2699. The comprehensive review of first-generation architectures and their rate budgets.
4. S.-K. Liao et al., Satellite-to-ground quantum key distribution (Nature, 2017) — arXiv:1707.00542. The Micius satellite result.
5. Wikipedia, Quantum repeater — concise overview.
6. Government of India, National Quantum Mission — the mission document covering the communication pillar that funds Indian repeater and satellite QKD efforts.