⚛️ quantum physics

Experimental Quantum Bernoulli Factories via Bell-Basis Measurements

This paper experimentally demonstrates an entanglement-assisted quantum Bernoulli factory on IBM superconducting hardware, utilizing Bell-basis measurements to realize classically inconstructible functions like $f(p)=2p$ and $f(p)=4p(1-p)$ , thereby validating the potential of quantum resources for enhanced stochastic simulation.

Original authors: Tanay Roy

Published 2026-02-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tanay Roy

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 have a slightly unfair coin. You don't know exactly how unfair it is; maybe it lands on heads 30% of the time, maybe 80%, but you don't know the number. In the world of math and computing, this is called a "p-coin."

For a long time, scientists have asked: Can we use this mysterious, unfair coin to create a perfectly fair coin (50/50) or other specific types of randomness, without ever figuring out what the original unfairness was?

This is the "Bernoulli Factory" problem.

The Classical Struggle: The "Keep Trying" Game

In the old, classical way (using just regular coins and math), if you want to make a fair coin from an unfair one, you have to play a game of "keep trying."

The Analogy: Imagine you flip your unfair coin twice. If you get "Heads then Tails," you call that a "Heads" result. If you get "Tails then Heads," you call that a "Tails" result. But if you get "Heads-Heads" or "Tails-Tails," you have to throw those results away and start over.
The Problem: If your coin is very unfair (say, 99% Heads), you will almost always get "Heads-Heads." You will have to throw away thousands of flips before you finally get a usable pair. It's like trying to find a needle in a haystack, but the haystack keeps growing the more you look.

The Quantum Solution: The "Magic Pair"

This paper, by Tanay Roy at Fermilab, shows how to solve this problem using quantum mechanics on a real computer (IBM's superconducting chips).

Instead of flipping coins one by one, the researchers prepared two "quantum coins" (called quoins) that were identical twins. They didn't just flip them; they linked them together using a quantum phenomenon called entanglement.

Think of entanglement like a pair of magic dice that are connected by an invisible string. Even though they are separate, what happens to one instantly affects the other.

The Bell-Basis Measurement: The "Magic Filter"

The researchers performed a special measurement called a Bell-basis measurement.

The Analogy: Imagine you have two coins. Instead of looking at them individually to see if they are Heads or Tails, you put them into a special "magic box" (the Bell measurement).
This box doesn't tell you "Heads" or "Tails." Instead, it sorts the pair into four specific categories (like sorting socks into pairs of matching colors).
The Magic Result: No matter how unfair the original coins were, this "magic box" sorts them in a way that guarantees a perfectly fair 50/50 outcome for one of the categories.

What Did They Actually Build?

Using this "magic box" approach on real hardware, the team demonstrated three things that are impossible or incredibly difficult to do with classical coins:

The Perfect Fair Coin: They turned their unknown unfair coins into a perfect 50/50 coin.
- Why it's cool: In the classical world, if your coin is almost always Heads, you need thousands of flips to get a fair result. In their quantum experiment, they only needed two quantum coins, no matter how unfair the original ones were. It's a constant, efficient cost.
The "Bernoulli Doubler": They created a function that doubles the probability of the coin (up to a limit).
- The Analogy: If your coin was 10% Heads, this machine turned it into a 20% Heads coin. If it was 40%, it became 80%.
- Why it's cool: Classical math says you cannot build a machine that does this perfectly without knowing the original number. The quantum machine did it anyway.
The "4p(1-p)" Function: They created a third type of coin that behaves in a specific, curved way (highest probability when the original coin is 50/50, and zero probability if the original is 0% or 100%).
- Why it's cool: This is another function that classical rules say is impossible to build exactly. The quantum machine built it using the same "magic box" data.

The Big Picture

The paper claims that by using entanglement and Bell measurements, they created a simple, efficient tool to process randomness.

Efficiency: They didn't need to guess the bias or throw away thousands of tries. They used a fixed, small number of quantum coins (2 for the fair coin, 4 for the others) every single time.
Self-Contained: They didn't need any outside help or extra random numbers. The "magic box" generated all the necessary randomness internally.
Real-World Test: They didn't just do this on paper; they ran it on actual IBM quantum computers. They found that while the computer's "noise" (glitches) made the results slightly imperfect, the core idea worked exactly as predicted.

In short, they showed that by linking two quantum coins together and looking at them as a pair, you can perform "magic" tricks with randomness that are strictly forbidden in the classical world.

Technical Summary: Experimental Quantum Bernoulli Factories via Bell-Basis Measurements

Problem Statement
The Bernoulli factory problem addresses the transformation of a source of imperfect randomness (a coin with unknown bias $p$ , or a $p$ -coin) into a new coin with a specific bias $f(p)$ using only a finite, random number of samples, without estimating $p$ . In the classical domain (Classical Bernoulli Factories or CBFs), the set of exactly constructible functions is severely constrained. Specifically, functions that approach 0 or 1 too rapidly or attain the values $\{0, 1\}$ within the interior of the domain ( $p \in (0,1)$ ) are not constructible by any algorithm that halts with probability one. Furthermore, even for constructible functions, the expected sample complexity can be large and strongly dependent on $p$ , diverging as $p$ approaches 0 or 1.

Methodology
This work presents an experimental implementation of a Quantum Bernoulli Factory (QBF) on IBM superconducting hardware. The protocol utilizes two identical input "quoins" (quantum coins), where the bias $p$ is encoded coherently in the qubit state $|p\rangle = \sqrt{1-p}|0\rangle + \sqrt{p}|1\rangle$ .

The core of the methodology involves performing a single Bell-basis measurement on the two-qubit system $|p\rangle \otimes |p\rangle$ . The Bell basis is defined as:
$|\Phi^\pm\rangle = \frac{|00\rangle \pm |11\rangle}{\sqrt{2}}, \quad |\Psi^\pm\rangle = \frac{|01\rangle \pm |10\rangle}{\sqrt{2}}$
When the input state is expanded in this basis, the probability amplitudes reveal specific statistical outcomes:

The probability of observing $|\Phi^+\rangle$ is exactly $1/2$ , independent of $p$ .
The conditional probability of observing $|\Psi^+\rangle$ given the outcome is in the set $\{|\Phi^-\rangle, |\Psi^+\rangle\}$ yields the function $f(p) = 4p(1-p)$ .
The statistics also allow for the realization of the "Bernoulli doubling" function $f(p) = 2p$ (for $p \leq 1/2$ ) and $2(1-p)$ (for $p > 1/2$ ).

The experimental circuit prepares two qubits via $Y(\theta)$ rotations where $\theta = 2\sin^{-1}\sqrt{p}$ , applies a CNOT gate followed by a Hadamard gate to transform the computational basis to the Bell basis, and performs standard measurements. The outcomes are mapped to the corresponding Bell states to generate the desired output coins.

Key Contributions and Results
The paper reports three primary experimental achievements:

Realization of Classically Inconstructible Functions: The authors successfully demonstrate the generation of the "Bernoulli doubling" primitive $f(p) = 2p$ (and its symmetric counterpart) and the function $f(p) = 4p(1-p)$ . These functions are impossible to construct exactly using classical algorithms that halt with probability one because they attain the value 1 at $p=1/2$ .
Internal Fair Coin Generation: Unlike previous approaches that may require external classical randomness sources for subroutines, this experiment generates an exact fair coin ( $f(p) = 1/2$ ) directly from the Bell-measurement statistics. The probability of observing $|\Phi^+\rangle$ is exactly $1/2$ regardless of the input bias $p$ .
Constant Resource Efficiency: The experiment demonstrates that the average number of input quoins required to generate these outputs is constant and independent of $p$ . Specifically, the fair coin requires two quoins, and the $f(p) = 4p(1-p)$ function requires four quoins on average. This contrasts sharply with classical methods (e.g., the von Neumann scheme), where the number of required coins diverges as $p \to 0$ or $p \to 1$ .

The authors benchmark the measured output biases against ideal theoretical predictions, observing deviations attributed to device noise inherent in the superconducting hardware.

Significance
The paper establishes a simple, resource-efficient experimental primitive for quantum-to-classical randomness processing. By utilizing entanglement and Bell-basis measurements, the work demonstrates that quantum resources can outperform classical ones in generating specific random variables that are strictly impossible to construct classically. The authors posit that these results support the viability of quantum Bernoulli factories for future quantum-enhanced stochastic simulation and sampling tasks. The approach is distinct from prior demonstrations by generating necessary fair coins internally via the same measurement statistics, thereby avoiding the need for additional external randomness sources.

The Classical Struggle: The "Keep Trying" Game

The Quantum Solution: The "Magic Pair"

The Bell-Basis Measurement: The "Magic Filter"

What Did They Actually Build?

The Big Picture

More like this