The Inverse Born Rule Equivalence. On the Informational Limits of Real-Valued Amplitude Encodings and the Measurement of Quantum Advantage in Data Embeddings

This paper proves that quantum data encodings restricted to real-valued amplitudes are mathematically equivalent to classical quadratic forms due to the absence of complex-phase interference, thereby establishing that genuine quantum advantage strictly requires complex structures and identifying the misinterpretation of real-amplitude models as quantum power as the "Inverse Born Rule Fallacy."

The Gist

The Big Question: Is the Quantum Computer Actually Doing Something New?

Imagine you have a new, super-fast kitchen appliance (a quantum computer) and you want to cook a complex meal (solve a data problem). You put your ingredients (data) into the machine. The big question this paper asks is: Is this machine actually cooking a new kind of dish, or is it just re-packaging a dish you could have made in a regular kitchen?

The authors, a team of researchers, discovered that a very popular way of feeding data into quantum computers is actually a "trap." It looks quantum, but mathematically, it's just a fancy version of a standard classical computer trick. They call this the "Inverse Born Rule Fallacy."

The "Real-Valued" Trap

In quantum computing, data is usually stored as "amplitudes" (numbers that determine the probability of a result). These numbers can be real (like 0.5 or -0.3) or complex (numbers that include an imaginary part, like $0.5 + 0.2i$).

The paper focuses on a specific method called Real-Valued Amplitude Encoding (and a sub-type called "Probability Loading").

• The Analogy: Imagine you are painting a picture.
  • Complex Encoding: You have a full palette of colors, including special "phase" pigments that can shift and interfere with each other to create new, shimmering effects.
  • Real-Valued Encoding: You are forced to use only black and white paint. You can mix them to make greys, but you can never create a new color or a shimmering effect.

The authors prove that if you only use "black and white" (real numbers) to load your data, no matter how you twist the knobs on your quantum machine later, the final result is mathematically identical to a simple classical quadratic form.

What does that mean?
It means the quantum computer isn't doing anything "quantum" here. It's just calculating a weighted sum of your data, exactly like a standard computer program could do in a few seconds. The "quantum advantage" (the speed or power boost) disappears.

The Secret Ingredient: The "Berry Connection"

Why does using only real numbers kill the quantum advantage? The authors found the geometric reason.

• The Analogy: Think of the data as a traveler walking on a map.
  • In a Complex system, as the traveler moves, they can spin around (change phase) in ways that are invisible to a simple observer but change the final destination. This spinning is called the Berry Connection. It's like a hidden compass that allows the traveler to take shortcuts through a "quantum tunnel."
  • In a Real-Valued system, the traveler is stuck on a flat, 2D sheet of paper. They can move forward and backward, but they cannot spin or twist. The "hidden compass" (Berry Connection) is broken; it reads zero.

Because the "spin" is gone, the quantum landscape collapses into a flat, classical landscape. The paper shows that for these real-valued methods, the complex geometry of quantum mechanics shrinks down to the boring, flat geometry of classical statistics.

How to Tell the Difference: The "Quantumness" Test

Since not all quantum methods are traps, the authors created a "diagnostic kit" to test if a method is actually quantum or just pretending. They use three main checks:

1. Phase Complexity (C): Does the data have "imaginary" parts? If $C=0$, it's just a classical trick. If $C>0$, it has real quantum potential.
2. The Berry Connection (|A|): Is there that hidden "spin" or rotation? If it's zero, the quantum advantage is dead.
3. Mutual Information (I): Are the different parts of the system tangled together (entangled)?

The Result of the Test:

• Probability Loading (The Trap): Fails all checks. It has no phase complexity and no Berry connection. It is mathematically identical to a classical kernel machine.
• Sandwich/Hamiltonian Encoding (The Real Deal): Passes the tests. They have complex phases and non-zero Berry connections. They can actually do things classical computers can't.

The Two Ways to Escape the Trap

The paper concludes that if you want a real quantum advantage (Type B), you must break the rules of the "Real-Valued Trap" in one of two ways:

1. Route 1: Use Complex Phases.

   • Analogy: Stop using just black and white paint. Start using the full color palette.
   • Method: Use encodings like "Sandwich" or "Hamiltonian" that introduce complex numbers. This creates the "hidden spin" (Berry connection) needed for true quantum interference.
   • Result: In their experiments, this method solved a tricky "XOR" puzzle perfectly, while the real-valued methods failed miserably.

2. Route 2: Re-upload the Data.

   • Analogy: If you are stuck with black and white paint, you can still make a masterpiece if you paint over the same canvas many times, layering the strokes.
   • Method: Instead of putting the data in once, you feed it into the circuit multiple times (Data Re-uploading).
   • Result: Even with real numbers, doing this many times creates complex patterns that a single layer couldn't. This allowed a "classical" encoding to solve hard problems, but only because the circuit depth (the number of layers) compensated for the lack of quantum phases.

The Bottom Line

The paper warns researchers: Don't be fooled by the label "Quantum."

If you are using Real-Valued Amplitude Encoding (or Probability Loading) with a standard setup, you are not getting a quantum advantage. You are just running a classical algorithm on a quantum machine. To get a genuine advantage, you must either use complex phases (the "colorful" route) or re-upload your data many times (the "layered" route).

The choice of how you put data into the machine is the most important decision in Quantum Machine Learning. If you choose the "Real-Valued" route, you are building a very expensive classical computer, not a quantum one.

Technical Summary

Technical Summary: The Inverse Born Rule Equivalence

Problem Statement
Quantum Machine Learning (QML) aims to leverage quantum interference to access hypothesis classes inaccessible to classical models. A critical design choice in QML pipelines is the data encoding (feature map), which embeds classical data into quantum states. Despite extensive research, there is no general criterion determining when a specific encoding provides a genuine "Type B" quantum advantage (an enlargement of the expressible hypothesis class) versus when it merely rephrases a classically solvable problem. This paper addresses the widespread use of real-valued amplitude encodings, including probability loading, and investigates whether they offer any intrinsic quantum advantage over classical kernel methods.

Methodology
The authors employ a rigorous theoretical framework combining linear algebra, information geometry, and differential geometry to analyze the hypothesis classes generated by parameterized quantum circuits (PQCs).

1. Theoretical Derivation: The paper proves an Equivalence Theorem demonstrating that any PQC model using real-valued amplitude encoding (where amplitudes $c_i \in \mathbb{R}$) composed with a data-independent unitary and computational-basis measurement yields a hypothesis class identical to that of a classical quadratic form.
2. Geometric Analysis: The authors analyze the differential geometry of the state manifold. They calculate the Berry connection ($A_\mu$) and the Fubini–Study metric for these encodings, comparing them to the classical Fisher–Rao metric.
3. Diagnostic Metrics: To operationalize these findings, three quantitative metrics are introduced to measure the "quantumness" of an encoding:
   • Phase Complexity ($C[\Phi]$): The magnitude of imaginary components in the state vector.
   • Berry Connection Magnitude ($|A_\mu|$): The measure of data-dependent phase rotation.
   • Mode-wise Mutual Information ($I[\Phi]$): The total entanglement across qubit modes.
4. Numerical Validation: The theoretical bounds are tested against five encoding strategies (Probability Loading, Angle Encoding, Data Re-uploading, Sandwich, and Hamiltonian) across four classification tasks (Two-moons, Concentric Circles, XOR, and Parity) using a state-vector simulator.

Key Contributions

• The Equivalence Theorem: The paper proves that for any real-valued amplitude encoding $|\psi_c\rangle = \sum c_i |i\rangle$ with $c_i \in \mathbb{R}$, the resulting hypothesis class is exactly the Reproducing Kernel Hilbert Space (RKHS) of a classical inner-product kernel $k(x, x') = (c(x)^\top c(x'))^2$. For probability loading, this reduces to the Hellinger affinity kernel. Consequently, these models provide zero Type B quantum advantage; they are classically realizable in $O(N^2)$ time.
• Geometric Mechanism (The Inverse Born Rule Fallacy): The authors identify the geometric cause of this collapse: restricting amplitudes to real numbers forces the Berry connection to vanish ($A_\mu \equiv 0$). This causes the quantum Fubini–Study metric to collapse directly onto the classical Fisher–Rao metric ($\Delta g = 0$). The paper terms the misidentification of such models as evidence of quantum power the "Inverse Born Rule Fallacy."
• Encoding Diagnostics: The paper introduces $C[\Phi]$, $|A_\mu|$, and $I[\Phi]$ as operational tools. It establishes that $C=0$ and $|A_\mu|=0$ are necessary conditions for the classical equivalence, while non-zero values indicate the presence of complex phases required for data-driven quantum interference.
• Identification of Escape Routes: The study delineates two distinct mechanisms to escape the classical bound established by the theorem:
  1. Complex Phases (Route 1): Using encodings where amplitudes are complex (e.g., Sandwich or Hamiltonian encoding), which generates a non-vanishing Berry connection.
  2. Data Re-uploading (Route 2): Using multi-layer circuits where data is injected repeatedly, breaking the assumption of a data-independent unitary and generating higher-order polynomial features.

Results

• Theoretical: The hypothesis class of any real-valued amplitude encoding PQC is strictly contained within the class of classical quadratic forms. The restriction to real values, not the $L_2$ normalization, is the mechanism that locks the model into the classical regime.
• Numerical:
  • Probability Loading (PL): As predicted, PL collapsed to near-chance performance on nonlinear tasks (Concentric Circles, XOR, Parity) where the decision boundary required non-quadratic or sign-flipping logic. It performed adequately only on the linearly separable Two-moons task.
  • Angle Encoding (AE): Despite having $C=0$ (purely real amplitudes), AE achieved high accuracy on nonlinear tasks (XOR, Circles). The authors clarify this does not violate the theorem; AE utilizes a trigonometric basis that naturally expresses finite Fourier series, effectively resolving periodic boundaries within the allowed quadratic form structure.
  • Data Re-uploading (RU): Achieved high accuracy on all tasks despite $C=0$, demonstrating that deep ansatz circuits can compensate for classical encodings by increasing the degree of the polynomial features.
  • Complex Encodings (SW, HE): Sandwich and Hamiltonian encodings, which possess $C>0$ and $|A_\mu|>0$, achieved perfect accuracy on the XOR task with fewer parameters than classical alternatives, validating the necessity of complex phases for specific structural advantages.

Significance and Claims
The paper asserts that the choice of feature map is the primary design decision in QML. Its significance lies in:

1. Demarcating the Boundary: It provides a strict, provable boundary between classical and quantum expressivity, showing that real-valued amplitude encodings (including probability loading) offer no intrinsic Type B advantage regardless of circuit depth or entanglement.
2. Correcting Misconceptions: It challenges the assumption that exponential data compression (a feature of amplitude encoding) equates to quantum advantage, showing instead that this compression comes at the cost of hypothesis class expressivity.
3. Operational Guidance: By introducing the "Inverse Born Rule Fallacy" and the diagnostic metrics ($C, |A_\mu|, \Delta g$), the paper offers a framework for researchers to audit encodings. It concludes that genuine quantum advantage requires either complex-phase structure in the encoding or data re-uploading to break the data-independent ansatz assumption. The paper emphasizes that quantum advantage is the product of encoding potential, data-encoding alignment, and circuit capacity, and that for real-valued encodings, the first factor is strictly zero.