⚛️ quantum physics

Lower bounds on non-local computation from controllable correlation

This paper introduces two new techniques based on controllable correlation and entanglement to establish lower bounds on the entanglement cost of non-local quantum computation for arbitrary unitaries, successfully resolving the cost for the CNOT gate and providing non-trivial bounds for other common two-qubit gates and Haar random unitaries.

Original authors: Richard Cleve, Alex May

Published 2026-04-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Richard Cleve, Alex May

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you and a friend are trying to perform a complex magic trick together, but you are in two completely different cities. You can't meet up, and you can't send each other physical props. All you have is a pre-shared "secret connection" (entanglement) and a single, simultaneous phone call to coordinate your moves.

This is the essence of Non-Local Quantum Computation (NLQC). The big question scientists have been asking is: How much "secret connection" (entanglement) do you actually need to pull off a specific trick?

For a long time, we knew how to do the tricks, but we didn't have a good way to prove how much "secret connection" was the absolute minimum required. It was like knowing you need some fuel to drive a car, but not knowing if you need a drop or a full tank.

This paper by Richard Cleve and Alex May introduces two new "fuel gauges" to measure exactly how much entanglement is needed for different quantum operations (gates).

The Two New "Fuel Gauges"

The authors developed two methods to calculate the minimum entanglement cost. Think of these as two different ways to test a machine's efficiency.

1. The "Remote Control" Gauge (Controllable Correlation)

Imagine you have a mysterious box (the quantum gate) that takes an input and gives an output. You have a "reference" system (a control panel) that is secretly linked to the input.

The Test: You ask your friend (who is holding the other side of the box) to press a button.
- If they press Button A, the connection between your control panel and the output stays strong and linked.
- If they press Button B, the connection gets scrambled or broken.
The Insight: If your friend can control whether the link stays strong or gets broken just by choosing a button, it proves that the "secret connection" (entanglement) between you two must have been real and substantial to begin with. If you had no entanglement, your friend couldn't influence your side of the link at all.
The Result: This method works for almost any quantum gate. It's a broad, reliable test that says, "Hey, this gate definitely needs some entanglement to work."

2. The "Shape-Shifter" Gauge (Controllable Entanglement)

This one is a bit more specific but gives a sharper, more precise answer for certain tricky gates (like the famous CNOT gate).

The Test: You start with a perfectly linked pair (maximally entangled). You ask your friend to choose an input.
- Scenario 1: They choose an input that keeps the link between you and your partner super strong (like a tight rope).
- Scenario 2: They choose a different input that makes the link completely disappear (like a loose string that falls apart).
The Insight: If your friend can switch the state of your connection from "super strong" to "completely gone" just by changing their input, it proves that the resource you shared was incredibly powerful.
The Result: For the CNOT gate (a fundamental building block of quantum computers), this gauge gave a perfect answer: You need exactly one "EPR pair" (the standard unit of quantum entanglement, like one pair of perfectly linked coins). Before this paper, no one could prove this was the minimum required; they just knew it was possible with one. Now, we know you can't do it with less.

Why Does This Matter?

Think of entanglement as a very expensive, rare currency.

Security: In quantum cryptography (like proving someone is really in a specific location), if we know the minimum "currency" needed to fake a signal, we can set security rules. If a hacker doesn't have enough entanglement, they can't cheat.
Physics & Gravity: Some theories suggest that the way space and time are connected (gravity) might be related to how much entanglement exists between particles. Understanding the "cost" of these connections helps physicists understand the universe.
Efficiency: If we know the exact cost of a gate, we can build quantum computers that don't waste resources.

The Big Takeaway

Before this paper, we were guessing how much "fuel" (entanglement) different quantum gates needed. We knew some needed a lot, and some needed none (like the SWAP gate, which is like just swapping seats without needing a secret link).

Now, the authors have built a calculator that can tell you the minimum fuel cost for almost any gate you throw at it.

For the CNOT gate, they solved the mystery: It costs exactly 1 unit.
For random gates, they found that almost all of them need some entanglement to work.
They also showed that if you need to do a trick twice, you need twice the fuel (a property called "parallel repetition").

In short, they turned a vague question ("How much entanglement is needed?") into a precise calculation, giving us a better map for navigating the strange world of quantum mechanics.

1. Problem Statement

The paper addresses the fundamental problem of determining the entanglement cost required to implement Non-Local Quantum Computation (NLQC).

Context: In NLQC, two parties (Alice and Bob) share entanglement and perform local operations with a single round of simultaneous communication to simulate a joint unitary operation $U_{AB}$ on their respective systems without physically bringing the systems together.
The Gap: While upper bounds on entanglement cost are well-understood (e.g., via teleportation or Clifford+ $T$ decompositions), lower bounds are largely uncharacterized. Previous techniques were restricted to specific, narrowly defined cases (e.g., hidden basis tasks or specific classical input scenarios) and failed to provide general lower bounds for simple, common unitaries like the CNOT gate.
Goal: To develop general techniques that can evaluate lower bounds on entanglement cost for any arbitrary unitary $U_{AB}$ , specifically focusing on two-qubit gates.

2. Methodology

The authors introduce two new, related techniques based on the concept of controllability. Both techniques rely on a specific setup (Figure 3 in the paper) where one input system ( $A$ ) is entangled or correlated with a reference system ( $Q$ ), while the other input ( $B$ ) acts as a control. The core idea is to measure how much the choice of input on $B$ can alter the correlation or entanglement between $Q$ and $A$ after the unitary is applied.

Technique A: Controllable Correlation (CC)

Concept: This technique measures the variation in Mutual Information $I(Q:A)$ between the reference $Q$ and input $A$ based on the choice of input state on the control wire $B$ .
Definition: A unitary $U_{AB}$ $U_{A B}$ has $(\lambda_1, \lambda_2)$ $(λ_{1}, λ_{2})$ -controllable correlation if there exist states $\phi_1, \phi_2$ $ϕ_{1}, ϕ_{2}$ on $B$ $B$ and a correlated state $P_{QA}$ $P_{Q A}$ such that:
- $\lambda_1 = \max_{\phi_1} I(Q:A)$
- $\lambda_2 = \min_{\phi_2} I(Q:A)$
Lower Bound: If the difference $\Delta \lambda = \lambda_1 - \lambda_2$ is positive, the entanglement of formation ( $E_f$ ) of the resource state $\Psi_{LR}$ must satisfy:
$E_f(L:R) \geq \frac{\lambda_1 - \lambda_2}{2} - \text{error terms}$
Properties: This bound holds for any unitary and satisfies parallel repetition (repeating the gate $n$ times scales the lower bound linearly by $n$ ).

Technique B: Controllable Entanglement (CE)

Concept: This technique is more restrictive but can yield tighter bounds. It focuses on the Entanglement of Formation ( $E_f$ ) between $Q$ and $A$ .
Definition: A unitary has $(\lambda_1, \lambda_2)$ $(λ_{1}, λ_{2})$ -controllable entanglement if:
- For input $\phi_1$ , the state $U(\Psi^+_{QA} \otimes \phi_1)$ results in high entanglement $E_f(Q:A) = \lambda_1$ .
- For input $\phi_2$ , the state $U(\Psi^+_{QA} \otimes \phi_2)$ results in a state close to separable (trace distance $\lambda_2$ to the set of separable states).
Lower Bound: Under the condition that $\lambda_2 \leq 1$ , the bound is:
$E_f(L:R) \geq \frac{\lambda_1 - 2\lambda_1^{1/4}}{2} - \text{error terms}$
Key Distinction: This technique requires the reference and input to be maximally entangled initially. It provides a tight lower bound for the CNOT gate (resolving its cost exactly) but is harder to satisfy for other gates.

3. Key Contributions

General Lower Bound Framework: The authors provide the first techniques capable of evaluating lower bounds for any chosen unitary, moving beyond case-specific proofs.
Resolution of CNOT Entanglement Cost: For the CNOT gate, the Controllable Entanglement technique proves a tight lower bound of 1 EPR pair ( $E_f \geq 1$ ), matching the known upper bound. This fully resolves the entanglement cost of the CNOT gate in the NLQC setting.
New Bounds for Common Gates: The paper provides the first known lower bounds for a wide variety of standard two-qubit gates, including:
- DCNOT: $E_f \geq 0.5$
- $\sqrt{\text{SWAP}}$ : $E_f \geq 0.30$
- Berkeley B Gate: $E_f \geq 0.601$
- XX Interaction: $E_f \geq 0.5$
- Random Unitaries: For Haar-random two-qubit unitaries, the techniques typically yield non-trivial lower bounds (average $\approx 0.23$ ).
Parallel Repetition Properties: The authors prove that these lower bounds scale linearly with the number of parallel repetitions of the gate, which is crucial for complexity-theoretic applications.
Noise Robustness: The bounds are adapted to handle noisy implementations (approximate unitaries) with explicit error terms dependent on the diamond norm distance.

4. Results Summary

The authors performed numerical optimizations to evaluate these bounds for various gates. The results are summarized in the paper's Table 2 (reproduced conceptually below):

Gate	Lower Bound (CE)	Lower Bound (CC)	Reference State Used
CNOT	1 (Tight)	0.5	$\Psi^+$ or $\rho_{cc}$
DCNOT	0	0.5	$\Psi^+$ or $\rho_{cc}$
Berkeley B	0.601	0.5	$\rho_{cc}$
XX ( $\pi/4$ )	1	0.5	$\Psi^+$ or $\rho_{cc}$
iSWAP	0	0.5	$\Psi^+$ or $\rho_{cc}$
$\sqrt{\text{SWAP}}$	0	0.30	$\Psi^+$
Random Unitary	0	$\approx 0.23$	$\Psi^+$ or $\rho_{cc}$

Note: For many gates (like CNOT), the Controllable Correlation (CC) bound is non-zero, while Controllable Entanglement (CE) is zero or harder to compute. However, for CNOT, CE provides the tightest possible bound.

5. Significance and Implications

Quantum Complexity & Cryptography: Understanding entanglement costs is vital for Quantum Position Verification (QPV). Lower bounds on entanglement are necessary to prove the security of QPV schemes against attackers with limited entanglement. This work provides tools to establish these security guarantees for a broader class of protocols.
Quantum Gravity: The paper notes connections to the AdS/CFT correspondence and wormholes, where non-local computation models are used to study spacetime geometry.
Faithfulness of Measures: The observation that Haar-random unitaries almost always have non-zero entanglement cost suggests that "zero-cost" unitaries (like SWAP or Identity) are a set of measure zero. This implies that entanglement is a generic resource for non-local computation.
Limitations: The current techniques apply strictly to unitary operations. Extending these bounds to general quantum channels remains an open problem, as the authors note that certain channels can mimic unitary correlation properties while requiring zero entanglement.

In conclusion, this paper breaks new ground by providing a systematic, computable method to determine the minimum entanglement required for non-local quantum computation, successfully resolving the cost for the CNOT gate and establishing non-trivial costs for a wide array of quantum logic gates.