Experimental quantum state learning with pairs of photons

This paper experimentally demonstrates a protocol for uniquely identifying the constituent pure states and their weights of a two-state qubit mixture by measuring single photons and retrospectively pairing them based on time-of-arrival, achieving high-fidelity discrimination between distinct preparations of the same mixed state with approximately 10,000 photons.

The Gist

Imagine you are trying to guess the secret recipe of a mysterious soup. You are allowed to taste the soup, but there's a catch: the soup is a perfect blend of two different broths (let's call them "Tomato" and "Basil").

If you just take a spoonful and taste it, you can tell it's a mix. You can measure how much tomato flavor there is versus basil. But you can't be 100% sure exactly which specific tomato or basil was used, because many different combinations of ingredients can create the exact same taste. In the world of quantum physics, this is called a "density matrix." It tells you the statistics of the mix, but it hides the identity of the individual ingredients.

The "Pairing" Trick
This paper describes a clever experiment where the scientists found a way to identify the exact ingredients, even though they are mixed together.

Here is the analogy:
Imagine Alice is sending Bob a stream of soup spoons. She promises that every spoon contains either pure "Tomato" or pure "Basil," but she mixes them up randomly.

• The Problem: If Bob just tastes the spoons one by one, he can only figure out the ratio (e.g., 50% tomato, 50% basil). He can't know if the "Tomato" is from a specific vine or the "Basil" from a specific garden.
• The Solution: Alice has a secret. She knows that every "Tomato" spoon she sends is secretly paired with another "Tomato" spoon, and every "Basil" spoon is paired with another "Basil" spoon. She doesn't tell Bob this before he tastes them. She just sends the spoons.
• The Magic Step: After Bob has tasted and recorded all the spoons, Alice sends him a list saying, "Okay, Spoon #1 and Spoon #42 were a pair. Spoon #5 and Spoon #99 were a pair."

By grouping the spoons into pairs after the fact, Bob can look at the data differently. Instead of seeing a blurry mix, he can now see that "When Spoon #1 was Tomato, its partner Spoon #42 was also Tomato." This extra layer of information allows him to mathematically separate the two ingredients and identify exactly what the "Tomato" and "Basil" states are, along with their exact probabilities.

What They Did in the Lab
The scientists didn't use soup; they used photons (particles of light).

1. The Source: They created pairs of photons using a special crystal.
2. The Mix: They manipulated the polarization (the direction the light waves vibrate) of the photons to create a random mix of two specific states (like vertical and horizontal vibration).
3. The Measurement: They measured the photons one by one, recording exactly when each one arrived.
4. The Pairing: Later, they used the arrival times to "pair up" the photons, just like Alice sending the list to Bob.
5. The Result: Using this "paired data," they successfully figured out the exact identity of the two hidden states and how often they appeared.

How Good Was It?
The team tested how close the two states could be before they became impossible to tell apart.

• They found that if the two states are too similar (like two shades of red that are almost identical), they need a lot of data to tell them apart.
• They discovered that with about 10,000 pairs of photons, they could identify the states with 99.99% accuracy.
• They also found a limit: if the two states are less than 15 degrees apart on the "color wheel" of light, the method can no longer distinguish them reliably.

Why Does This Matter?
The paper shows that by using "time-of-arrival" information to group particles into pairs after they have been measured, we can learn more about a quantum system than we thought possible with standard measurements. It's like being able to solve a puzzle by looking at the pieces twice: once individually, and then again after you've been told which pieces belong together.

The researchers also explored how much information could be packed into these mixtures. They found that while you can pack more "bits" of information by making the states very close together, it requires exponentially more photons to read them correctly. It's a trade-off: you can send a denser message, but you have to send a much larger volume of light to decode it.

In Summary
This experiment proves a theoretical idea: if you have a stream of quantum particles and you know which ones came in pairs, you can uniquely identify the specific states making up a mixture, something that is usually impossible with single-particle measurements alone. They did this with light, showing that "learning from pairs" is a real, working technique.

Technical Summary

Technical Summary: Experimental Quantum State Learning with Pairs of Photons

Problem Statement
Standard quantum state tomography allows for the estimation of a density matrix ($\rho$) describing an ensemble of identically prepared quantum systems. However, a fundamental limitation exists: a mixed-state density matrix generally does not possess a unique decomposition into its constituent pure states. Consequently, while an observer (Bob) can determine the statistical mixture $\rho_{single} = p_0 |\psi_0\rangle\langle\psi_0| + p_1 |\psi_1\rangle\langle\psi_1|$, he cannot uniquely identify the specific pure states $|\psi_0\rangle$ and $|\psi_1\rangle$ or their probabilities $p_0$ and $p_1$ from single-particle measurements alone.

Agarwal et al. [12] proposed a theoretical protocol to overcome this limitation. They demonstrated that if the qubits arrive in pairs where both members of a pair are in the same state, the resulting two-qubit density matrix, $\rho_{pair} = p_0 |\psi_0\psi_0\rangle\langle\psi_0\psi_0| + p_1 |\psi_1\psi_1\rangle\langle\psi_1\psi_1|$, admits a unique decomposition into pure product states. This allows an observer to infer the identity of the specific pure states and their weights. This paper presents an experimental demonstration of this "learning-from-pairs" concept using photon polarization.

Methodology
The experiment utilizes a spontaneous parametric down-conversion (SPDC) source to generate pairs of degenerate 808 nm photons. One photon (the idler) acts as a trigger, while the other (the signal) is sent through a polarization preparation stage. A liquid-crystal wave plate (LCWP) modulates the polarization of the signal photons based on applied voltage, switching between two settings to generate a statistical mixture of two distinct polarization states, $|\psi_0\rangle$ and $|\psi_1\rangle$, with probabilities $p_0$ and $p_1$.

The signal photons undergo single-photon polarization tomography, measuring projections in six bases (H/V, R/L, D/A). Crucially, the experiment records the time-of-arrival for every detected photon. Although the photons are measured individually, the time-stamp data is used post-measurement to "pair up" the photons based on the known generation statistics. This pairing allows the reconstruction of the two-photon density matrix $\rho_{pair}$ from the single-photon detection data. Following the protocol in Ref. [12], this $\rho_{pair}$ is then decomposed to uniquely identify $|\psi_0\rangle$, $|\psi_1\rangle$, $p_0$, and $p_1$.

Key Results
The authors investigated the protocol's performance under various conditions:

1. Fidelity and Sample Size: For orthogonal states (e.g., $|H\rangle$ and $|V\rangle$) with equal or unequal probabilities, the fidelity of the reconstructed states approaches 99.99% with approximately $10^4$ detected photon pairs. The fidelity scales as $1 - F \propto 1/N$, consistent with theoretical predictions.
2. Non-Orthogonal States: When the two states are non-orthogonal (separated by $\pi/2$ on the Poincaré sphere), more pairs are required to achieve the same fidelity. The protocol successfully distinguishes states separated by angles as small as $15^\circ$. Below this threshold, the curves for the two states overlap, indicating an inability to distinguish them with the given sample size.
3. State Discrimination: The experiment tested the ability to distinguish between two different preparations of the same single-qubit mixed state (where the underlying pure states differ). By varying the angle $\theta$ between the pure states in two different mixtures, the authors found that for $N \sim 10^4$, mixtures separated by angles less than $5^\circ$ become difficult to discriminate.
4. Information Capacity: The study calculated the number of distinct mixtures that can be encoded and distinguished. For a separation angle $\theta$, the number of distinguishable mixtures scales as $8/\theta^2$, yielding a bit capacity of $\log_2(8/\theta^2)$. While reducing $\theta$ increases the bits per symbol, the required photon count scales as $1/\theta^2$, leading to a decrease in bits per photon.

Significance and Claims
The paper claims to experimentally validate the "learning-from-pairs" protocol, demonstrating that access to pairing information (even if determined post-measurement via time-of-arrival) allows for the unique decomposition of a mixed state into its pure components. This effectively bypasses the non-uniqueness inherent in standard single-particle tomography.

The authors emphasize that this approach enables the inference of non-linear functions of the density matrix (specifically the decomposition into pure states) which are inaccessible via single-particle measurements alone. The work establishes a practical limit for this discrimination: with $\sim 10^4$ pairs, the system can resolve pure states separated by $15^\circ$ and distinguish between different mixtures of orthogonal states separated by less than $5^\circ$. The results suggest that while high information density is theoretically possible by reducing the separation angle, the resource cost (number of photons) increases quadratically, potentially favoring smaller symbol alphabets in communication settings.