End-to-End PDE-Based Quantum Algorithms for Multi-Asset Option Pricing under Local and Stochastic Volatility

This paper presents a complete end-to-end quantum algorithm for pricing multi-asset European options under local and stochastic volatility models, demonstrating polynomial speedups over classical finite-difference methods in terms of gate complexity while providing explicit resource accounting and numerical benchmarks.

The Gist

Imagine you are a chef trying to predict the price of a complex dish (an "option") that depends on the future prices of several ingredients (assets like stocks). The price of this dish isn't just a simple average; it's influenced by how volatile (jumpy) the ingredients are and how they move in relation to each other.

In the financial world, calculating this price is like solving a massive, multi-dimensional maze called a Partial Differential Equation (PDE).

The Problem: The "Curse of the Grid"

Traditionally, to solve this maze, computers use a method called Finite Differences. Imagine you have to map out a 3D city to find a specific address.

• The Classical Approach: You lay down a grid of streets. If you have 1 ingredient, you need a 1D line of grid points. If you have 10 ingredients, you need a 10D hyper-grid.
• The Bottleneck: As you add more ingredients (assets), the number of grid points explodes exponentially. It's like trying to fill a room with sand; if you double the number of ingredients, the amount of sand (computing power) needed doesn't just double—it multiplies by a huge factor. This is known as the "curse of dimensionality." For complex dishes with many ingredients, classical computers get stuck in the sand.

The Solution: A Quantum "Magic Lens"

This paper proposes a new way to solve this problem using Quantum Computers. Instead of building a giant physical grid of sand, the authors developed an "end-to-end" quantum pipeline that acts like a magic lens.

Here is how their system works, step-by-step:

1. The Setup (State Preparation)
First, the computer takes the "recipe" (the contract details, strike prices, and market data) and encodes it into a quantum state. Think of this as loading the initial ingredients into a quantum blender. They use a clever trick called Schrödingerization to turn the messy, non-quantum math of the pricing equation into a format that a quantum computer can understand (a "unitary" evolution).

2. The Journey (Quantum Evolution)
Instead of walking through every single grid point one by one (like a classical computer), the quantum computer evolves the entire system at once. It's like dropping a stone in a pond and watching the ripples spread out instantly across the whole surface, rather than measuring the water level at every single point individually. The paper uses advanced techniques (like Hamiltonian simulation) to let the quantum state "flow" backward in time from the future (maturity) to the present.

3. The Reveal (Readout)
Once the quantum state has evolved, the computer needs to tell us the price. Since we can't look at the whole quantum soup at once, the authors use a technique called Amplitude Estimation. This is like taking a single, highly precise sample from the soup to estimate the flavor of the whole pot. They specifically look for the price at a specific point (the current market state).

The Results: A Speed Boost

The authors tested this on two famous financial models:

• Black-Scholes: A standard model for pricing options.
• Heston: A more complex model that accounts for "volatility smiles" (the fact that market volatility isn't constant; it changes based on the price, creating a smile-shaped curve).

The Findings:

• Polynomial Speedup: For a dish with $d$ ingredients and a grid size of $N$, the classical computer takes time proportional to $N^{d+2}$. The quantum algorithm reduces this to roughly $N^{d/2 + 2}$ (for Black-Scholes) or $N^{d+2}$ but with a much smaller exponent in the leading term.
• The Analogy: If the classical computer has to count every grain of sand in a beach, the quantum computer can estimate the volume by looking at a much smaller, representative sample, saving a massive amount of time as the beach gets bigger.
• Real-World Validation: The paper didn't just do math on paper. They ran simulations and showed that their quantum method successfully recreated the "volatility smile" (the curved graph of implied volatility) just as well as classical methods, proving it captures the real market behavior.

Important Caveats (The Fine Print)

The authors are very careful to state what this doesn't do yet:

• It's not a magic wand for everything: The speedup is significant, but it doesn't completely eliminate the "curse of dimensionality." The cost still grows as you add more assets, just much slower than before.
• It's theoretical right now: The "gate complexity" (the number of steps) is calculated for a perfect, error-free quantum computer. Real quantum computers today are noisy and small.
• Specific Scope: This method works best for European-style options (where you can only exercise at the end) and specific types of multi-asset contracts. It doesn't yet handle every possible exotic financial derivative (like those with early exercise features).

Summary

In simple terms, this paper builds a complete, theoretical "quantum assembly line" for pricing complex financial options. It takes classical data, runs it through a quantum engine that simulates the future price movements of multiple assets simultaneously, and outputs a price. The result is a method that is mathematically proven to be significantly faster than current classical methods for high-dimensional problems, successfully reproducing complex market patterns like the "volatility smile."

Technical Summary

Problem Statement
The pricing of multi-asset options under local-volatility (Black–Scholes) and stochastic-volatility (Heston) models naturally leads to high-dimensional parabolic partial differential equations (PDEs). Classical numerical methods for solving these PDEs, such as finite-difference schemes or matrix-exponential methods, suffer from the "curse of dimensionality," where computational cost grows exponentially with the number of assets ($d$) and the grid resolution ($N$). Specifically, for a grid with $N$ points per spatial direction, classical grid-based methods scale as $\tilde{O}(N^{d+2})$ for Black–Scholes and $\tilde{O}(N^{2d+2})$ for Heston (where the Heston model includes variance variables). While quantum algorithms have been proposed for option pricing, many rely on Monte Carlo path simulation or variational approaches that face challenges with state preparation, normalization, and readout, particularly for complex, high-dimensional PDEs with specific boundary conditions.

Methodology
The authors develop an end-to-end quantum PDE framework that takes classical contract and model data as input and returns classical estimates of option values. The workflow consists of three main stages:

1. Discretization and Formulation: The pricing PDEs are discretized in space using second-order finite differences, transforming the continuous PDE into a sparse, finite-dimensional system of ordinary differential equations (ODEs). To handle non-homogeneous boundary conditions and source terms, the affine ODE system is augmented into a homogeneous linear system by adding an auxiliary state variable.
2. Quantum Evolution (Schrödingerisation): Since the resulting ODE system is generally non-unitary (due to the non-Hermitian nature of the pricing operator), the framework employs Schrödingerisation. This technique embeds the non-unitary dynamics into a larger unitary system by introducing an auxiliary continuous variable ($\xi$) and mapping the dynamics to a Schrödinger equation. The system is then discretized in the $\xi$ dimension and transformed via Fourier methods to yield a Hermitian Hamiltonian.
3. State Preparation and Readout:
   • State Preparation: The payoff vector (e.g., Worst-of Call) and the smooth cut-off profile required for Schrödingerisation are prepared as quantum states using block-encodings of coordinate operators and Quantum Singular Value Transformation (QSVT).
   • Evolution: The discretized pricing dynamics are simulated via Hamiltonian evolution using block-encoding and alternating phase modulation sequences (Quantum Signal Processing).
   • Readout: Selected option values are recovered from the final quantum state. Because the desired price is encoded in a single amplitude, the framework utilizes Quantum Amplitude Estimation (QAE) combined with a Hadamard test. Crucially, the readout overhead scales with the square root of the grid volume ($2^{W/2}$), where $W$ is the number of system qubits.

Key Contributions

• End-to-End Pipeline: The paper provides a complete, explicit pipeline from classical input data to classical output, including detailed resource accounting for state preparation, Hamiltonian simulation, normalization recovery, and readout.
• Complexity Analysis: The authors derive the leading gate complexity for single-point option price recovery:
  • Multi-asset Black–Scholes: $\tilde{O}(d^2 N^{2 + d/2})$.
  • Multi-asset Heston: $\tilde{O}(d^2 N^{d + 2})$.
  • Compared to classical grid-based baselines ($\tilde{O}(N^{d+2})$ and $\tilde{O}(N^{2d+2})$ respectively), the quantum framework offers polynomial improvement factors of $N^{d/2}$ and $N^d$.
• Numerical Validation: The framework is numerically validated against analytical solutions (Black–Scholes), semi-analytical benchmarks (Heston characteristic function), and classical numerical solvers (finite-difference and matrix-exponential).
• Implied Volatility Recovery: In the Heston setting, the framework successfully recovers option prices across a range of strikes and reconstructs the associated non-flat implied-volatility smile/skew, demonstrating its ability to capture market-relevant stochastic volatility features.

Results

• Theoretical Performance: The analysis confirms a polynomial quantum advantage in the grid size $N$ for both Black–Scholes and Heston models. The advantage becomes more pronounced as the number of assets $d$ increases, though the exponential dependence on $d$ (the curse of dimensionality) is mitigated but not eliminated.
• Numerical Accuracy: Simulations show that the quantum framework produces option prices consistent with classical and semi-analytical benchmarks. In the Heston case, the recovered implied volatility surfaces exhibit the correct skew structure (e.g., equity skew) and match the semi-analytical SSVI fits with relative discrepancies below 1%.
• Error Sources: The primary errors in the numerical results are attributed to coarse spatial discretization and domain truncation, rather than the quantum simulation itself. The additional error introduced by the Schrödingerisation procedure is found to be subleading.

Significance and Claims
The paper claims to provide the first demonstration of a quantum PDE-based pricing workflow that successfully recovers an implied-volatility smile/skew from quantum-computed option prices. It emphasizes that the framework offers explicit performance guarantees and a complete resource analysis, distinguishing it from variational or heuristic approaches.

The authors maintain a modest tone regarding the practical advantage:

• They clarify that the comparison is at the level of asymptotic scaling rather than wall-clock time, noting that logical gate counts do not directly translate to physical execution time without considering error correction overheads.
• They acknowledge that the "curse of dimensionality" is not eliminated; the cost still grows exponentially with the number of assets, but the growth rate is reduced (from $N^d$ to $N^{d/2}$ for Black–Scholes).
• They note limitations, such as the current focus on European-style payoffs with boundary conditions compatible with the framework (e.g., Worst-of Call), and the exclusion of path-dependent or early-exercise features which would require additional state augmentation or iterative solving layers.

Overall, the work establishes a rigorous theoretical and numerical foundation for using quantum algorithms to solve high-dimensional financial PDEs, offering a viable alternative to classical grid-based methods for multi-asset pricing.