Squeezed-state semi-device-independent quantum randomness generation

This paper derives a closed-form expression for the certified asymptotic Shannon rate in semi-device-independent quantum randomness generation using a trusted binary pure-state source and an untrusted binary detector, demonstrating that including deterministic extreme points in the POVM optimization yields a correct, lower bound and applying this result to analyze the impact of squeezing on squeezed-coherent BPSK sources.

The Gist

The Big Picture: Making True Random Numbers

Imagine you are running a casino. You need a machine that generates truly random numbers (like rolling a die) to ensure the games are fair. In the quantum world, we use light particles (photons) to make these numbers because, unlike a weighted die, quantum particles are fundamentally unpredictable.

However, there's a catch: Do you trust the machine?

• Fully Trusted: You built the machine yourself, checked every screw, and know exactly how it works. (Very fast, but if you made a mistake, the numbers aren't random).
• Fully Untrusted: You bought a "black box" from a stranger. You have no idea what's inside. (Very secure, but the machine is so slow it's useless for real life).

This paper focuses on the "Semi-Device-Independent" middle ground. It's like trusting the ingredients you put into a cake (the light source) but not trusting the oven (the detector) that bakes it. The oven might be broken, or it might be secretly rigged by a hacker. The goal is to prove that even with a suspicious oven, the cake (the random numbers) is still safe to eat, provided you know the ingredients well.

The Problem: The "Perfect" Math Was Wrong

The authors looked at a specific type of quantum randomness generator using squeezed light (a special state of light that is "squished" to make it more predictable in one way and less in another).

They found a major error in how scientists had been calculating the safety of these machines for years.

• The Old Way: Scientists used a formula that assumed the "oven" (detector) could only do two things: measure the light or ignore it. They ignored a third, sneaky possibility: the oven could just guess the answer every time without looking at the light at all.
• The Mistake: By ignoring this "lazy guess" option, the old math thought the machine was safer than it actually was. It was like calculating how hard it is to break a bank vault, but forgetting that a thief could just walk in through the unlocked back door.
• The Result: The old formula said you could get 0.25 bits of randomness. The new, correct formula says you only get 0.06 bits. That's a huge difference—like thinking you have a full wallet when you actually only have a few coins.

The Solution: A New "Safety Certificate"

The authors derived a new, closed-form formula (a single, neat equation) that accounts for all possible tricks a hacker could play, including the "lazy guess."

Think of this formula as a universal safety certificate.

1. Input: You tell the formula two things:
   • How similar the two light states are (the "overlap").
   • How often the detector makes a mistake (the "error rate").
2. Output: It spits out the exact amount of guaranteed randomness you can extract, no matter how the detector is rigged.

This formula is an "unconditional upper bound," meaning it is the absolute maximum randomness you can ever claim to have. If your machine performs better than this formula predicts, you are lying. If it matches, you are safe.

The Squeezing Trade-Off: The "Tightrope"

The paper then applies this new formula to squeezed light. Imagine squeezing a balloon.

• Squeezing more makes the balloon very thin in one direction (making the two light states very different and easy to tell apart).
• The Catch: While it makes them easier to tell apart, it also makes the "lazy guess" trick more effective for a hacker.

The authors discovered a trade-off:

• If you squeeze the light too much to make the states distinct, you actually lose certified randomness because the hacker can exploit the setup more easily.
• If you squeeze it too little, the states are too similar, and the machine can't tell them apart.

They found a "sweet spot" (or rather, the edges of the range) where you get the most randomness. Interestingly, the "perfect" squeezing for distinguishing the states (the usual goal in physics) is actually the worst spot for generating randomness.

The "Hacker" Model

The paper also clarifies who the "hacker" (adversary) is.

• The Scenario: The hacker controls the detector and has a secret notebook (classical side information) that tells them how the detector is behaving.
• The Limit: The paper proves that if the hacker is allowed to hold a "quantum purification" (a magical quantum notebook that tags every single outcome), they can steal all the randomness, reducing the guaranteed rate to zero.
• The Assumption: This paper assumes the hacker's notebook is classical (just a list of numbers), not quantum. This is a specific, realistic assumption that allows the math to work.

Summary

1. We fixed a math error: Previous calculations ignored a "lazy" hacking strategy, making quantum random number generators look safer than they were.
2. We have a new rule: A new formula gives the true maximum randomness you can get, accounting for all detector tricks.
3. Squeezing is tricky: In this specific setup, squeezing light to make it more distinct actually hurts your randomness guarantee. You have to balance the two carefully.
4. The result: This is the first time this specific type of "squeezed-state" generator has been analyzed with this level of security, providing a reliable "safety certificate" for building these devices.

Technical Summary

Technical Summary: Squeezed-state semi-device-independent quantum randomness generation

Problem Statement
The paper addresses the generation of certified quantum randomness in a semi-device-independent (semi-DI) framework. In this setting, the user possesses a trusted binary pure-state source but an untrusted binary detector. The adversary is assumed to hold classical side information (a hidden variable $\lambda$) that determines the detector's behavior. While fully device-independent protocols offer strong security, they are limited to low rates due to the need for loophole-free Bell violations. Conversely, fully trusted hardware offers high speeds but lacks security against untrusted components. Semi-DI protocols bridge this gap by relying on explicit physical assumptions, such as a dimension bound or a trusted state overlap.

A specific gap identified in the literature is the treatment of binary-qubit Positive Operator-Valued Measures (POVMs). Previous closed-form solutions for the certified Shannon rate in this model restricted optimization to strictly projective measurements. The authors argue that this restriction is insufficient because the convex set of binary-qubit POVMs includes two rank-deficient extreme points: the deterministic POVMs $M_0 = 0$ and $M_0 = I$. Omitting these deterministic components leads to an overestimation of the certified randomness, particularly in the rate-positive interior of the parameter space.

Methodology
The authors formulate the certification of asymptotic i.i.d. Shannon randomness as a linear programming (LP) problem over the convex set of classical-$\lambda$ binary POVMs acting on the two-dimensional subspace of the source states.

1. Model Definition: The source prepares states $|\psi_0\rangle$ and $|\psi_1\rangle$ with a trusted Gram overlap $\delta = \langle \psi_0 | \psi_1 \rangle$. The detector is modeled as a mixture of POVMs $\{M_{b|\lambda}\}$ indexed by an adversary-held variable $\lambda$. The observed symmetric error probability is $q$.
2. Extended Optimization: Unlike prior work, the optimization includes the full set of extreme points: rank-one projectors $\Pi_\theta$ and the deterministic endpoints $M_0=0$ and $M_0=I$. The objective is to minimize the conditional entropy $H_{cert}(\delta, q)$ subject to marginal constraints matching the observed error rate.
3. Closed-Form Derivation: The authors derive a closed-form expression for the certified rate, $H_{ext}^{Sh}(\delta, q)$, by constructing a primal solution (an explicit adversarial strategy) and a dual witness (an affine function bounding the entropy).
4. Application to Squeezed States: The derived formula is applied to squeezed-coherent Binary Phase-Shift Keying (BPSK) sources. The overlap $\delta$ and error probability $q$ are expressed in terms of the displacement amplitude $N$, squeezing parameter $r$, and detection efficiency $\eta$.

Key Contributions

• Correction of the Projective-Only Approximation: The paper demonstrates that restricting optimization to projective measurements significantly overestimates the certified rate. For example, at $\delta=0.5$ and $q=0.15$, the projective-only formula yields 0.25 bits, whereas the correct rate including deterministic components is only 0.06 bits (a factor-of-four correction).
• Closed-Form Rate Expression: The authors provide an unconditional upper bound on the certified asymptotic Shannon rate, $H_{ext}^{Sh}(\delta, q)$, which is tight on a numerically verified dual-feasibility region containing all operating points used in the study. The expression depends solely on the trusted overlap $\delta$ and the observed error $q$.
• Adversarial Model Clarification: The work clarifies that the side information is classical. It explicitly notes that if the adversary were allowed to hold a detector-purification register (quantum side information) that tags the outcome, the certified rate would drop to zero.
• Squeezing Trade-off Analysis: The paper analyzes squeezed-coherent BPSK sources, showing that while squeezing improves state distinguishability (reducing $\delta$), it can simultaneously reduce the certified randomness. This creates a trade-off where the rate-minimizing squeezing (which maximizes distinguishability) is often the worst choice for randomness generation.

Results

• Theoretical Bounds: The derived closed-form expression is proven to be an upper bound on the certified rate everywhere in the feasible region. It becomes an exact equality within the dual-feasibility region $R_{cert}$, which includes all operating points discussed in the paper.
• Squeezed-State Performance:
  • For a fixed mean photon number $\bar{n}$, the certified rate is not maximized at the squeezing level $r^*$ that maximizes state distinguishability. Instead, $r^*$ minimizes the certified rate.
  • The certified rate increases as the system moves away from $r^*$ toward the endpoints of the feasible squeezing range (either no squeezing, $r=0$, or maximum squeezing where the displacement amplitude $N \to 0$).
  • In the lossy regime ($\eta=0.7$), the benefit of squeezing is diminished, and the no-squeezing baseline often provides comparable or better certified rates depending on the constraints.
• Feasibility Regions: The paper defines strong-feasibility intervals for the error probability $q$ based on $\delta$. Outside the rate-positive interior, the certified rate drops to zero (a plateau), corresponding to decompositions where the adversary can simulate the observed statistics using zero-entropy deterministic strategies.

Significance and Claims
The authors claim this work provides the first semi-DI squeezed-state QRNG rate formula under a trust model where the source is trusted via Gram overlap and the detector is untrusted with classical side information. The primary significance lies in the rigorous inclusion of deterministic POVM components, which corrects previous overestimations of security. The paper positions this result as a necessary refinement for practical QRNGs using squeezed states, highlighting that optimizing for state distinguishability is not synonymous with optimizing for randomness generation. The authors explicitly state that the closed-form expression serves as a certified rate guarantee only within the numerically verified dual-feasibility region, while remaining a valid (though potentially loose) upper bound elsewhere.