⚛️ quantum physics

Trading Imaginary Time for Randomness in Ground State Preparation

This paper introduces Twirled Imaginary-Time Evolution (TITE), a method that pairs imaginary-time evolution with random real-time evolution to quadratically suppress ground state errors, thereby halving the required imaginary time and achieving a quadratic reduction in the computational cost of ground state preparation.

Original authors: Alvan Arulandu, John M. Martyn, Isaac L. Chuang

Published 2026-08-04

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Alvan Arulandu, John M. Martyn, Isaac L. Chuang

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 are trying to find the deepest, quietest valley in a vast, foggy mountain range. This is the kind of problem quantum computers are built to solve: finding the "ground state," or the lowest energy level, of a complex system like a molecule or a magnetic material. To do this, scientists use a clever trick called "imaginary-time evolution." Think of it as a magical filter that slowly drains the energy out of a system, letting the high-energy, bouncy states fade away until only the calm, low-energy ground state remains.

However, there's a catch. This magical filter is incredibly expensive to run. The longer you let it run to get a clearer picture, the more times you have to repeat the whole experiment to get a single good result. It's like trying to hear a whisper in a storm; the more you want to hear clearly, the more times you have to shout and listen again, until the cost becomes impossible. This paper tackles that expensive problem by introducing a new way to mix things up, using a bit of randomness to make the filter work much more efficiently without needing more expensive equipment.

The researchers, Alvan Arulandu, John M. Martyn, and Isaac L. Chuang, have developed a new method they call "Twirled Imaginary-Time Evolution" (TITE). Their big idea is to trade a little bit of the expensive, energy-draining "imaginary time" for a free, random spin of "real time."

Here is how it works in plain language: Imagine you are trying to tune a radio to a specific station (the ground state). The standard method involves slowly turning the dial (imaginary time) to filter out all the static and other stations. But this process is slow and requires you to keep hitting "reset" and trying again over and over because the signal gets weaker.

The authors realized that once you've turned the dial most of the way, you don't need to keep turning it slowly. Instead, you can take the radio and shake it randomly for a moment (real-time evolution). This shaking doesn't cost any extra energy or require extra resets because it's a natural, reversible process. By shaking the radio for a random amount of time, chosen from a specific pattern, the "static" (the unwanted excited states) gets scrambled and cancels itself out.

The paper proves that this random shaking is incredibly powerful. It shows that by replacing roughly half of the expensive imaginary time with this random real-time shaking, you can achieve a quadratic reduction in the total cost. Because the cost of the standard method grows exponentially with the time required (scaling as $e^\beta$ ), cutting the required imaginary time in half reduces the cost from $e^\beta$ to $e^{\beta/2}$ . While this remains an exponential function, it is a massive improvement: it effectively reduces the computational burden by the square root of the original cost. This means the error in the final result drops quadratically relative to the remaining imaginary time—meaning if you were off by a tiny bit before, the new method reduces that error to the square of that tiny bit, making the result significantly more accurate.

The authors tested this idea using computer simulations of a complex chain of magnetic spins (a non-integrable Ising chain). In these noisy simulations, their new TITE method showed substantial improvements over the standard way of doing things. They demonstrated that you can reach the same high-quality result with significantly less computational cost.

Crucially, the paper explains that this isn't just a magic trick for one specific type of computer; it works with any of the current methods scientists use to run imaginary-time evolution, whether they use step-by-step approximations or more advanced signal processing. The key insight is that while the energy of the system stays the same during the random shaking, the accuracy of the state improves because the random motion washes out the errors that usually plague these calculations.

So, the main finding is that you can drastically slash the cost of finding the ground state of a quantum system by adding a dash of randomness. The paper proves mathematically that this "temporal twirling" suppresses errors much faster than the old methods, and their simulations back up the theory, showing that this approach could make preparing quantum states much more practical for future quantum computers.

Technical Summary: Trading Imaginary Time for Randomness in Ground State Preparation

Problem Statement
Imaginary-time evolution (ITE) is a foundational method for preparing ground states of many-body Hamiltonians on quantum computers. Theoretically, applying the non-unitary operator $e^{-\beta H}$ to an initial state suppresses excited-state contributions exponentially with the imaginary time $\beta$ , converging to the ground state. However, practical implementations face a significant cost barrier: because $e^{-\beta H}$ is non-unitary, it reduces the norm of the wavefunction. Correcting this norm requires either post-selection or amplitude amplification, leading to a sample complexity and/or classical computational cost that scales exponentially with $\beta$ (i.e., $e^{O(\beta)}$ ).

Furthermore, a critical limitation exists in the convergence behavior of standard ITE. While the ground state energy converges rapidly (with error scaling as $O(\delta^2)$ where $\delta$ is the excited-state amplitude), the trace distance to the true ground state—and consequently the error in arbitrary observables—converges more slowly (scaling as $O(\delta)$ ). This discrepancy means that achieving high accuracy for general observables often requires significantly larger $\beta$ than what is suggested by energy convergence alone, exacerbating the exponential cost.

Methodology: Twirled Imaginary-Time Evolution (TITE)
The authors propose Twirled Imaginary-Time Evolution (TITE), a randomized algorithm designed to mitigate the cost of ground state preparation. The core methodology involves replacing a portion of the required imaginary time with a "temporal twirling" process using real-time evolution.

The algorithm proceeds in two stages:

Imaginary-Time Evolution: The system undergoes ITE for a reduced duration $\beta' \approx \beta/2$ . This brings the state within a trace distance of $O(\epsilon)$ of the ground state.
Temporal Twirling: The state is subsequently subjected to real-time evolution $e^{-iHt}$ for a random duration $t$ drawn from a carefully designed probability distribution $\mathcal{D}$ .

The theoretical foundation relies on the Campbell-Hastings Mixing Lemma, generalized here to quantum states. The lemma posits that if a set of unitaries approximates a target with error $a$ individually, but their average approximates the target with error $b$ , the resulting mixed state (channel) deviates from the target by an error of order $a^2 + 2b$ .

In TITE, the randomization over real-time durations $t$ causes the relative phases of the excited-state components to average out. If the distribution $\mathcal{D}$ is chosen such that its characteristic function decays sufficiently fast for frequencies $\omega \ge \Delta$ (the spectral gap), the average state achieves a trace distance error of $O(\epsilon^2)$ , a quadratic improvement over the $O(\epsilon)$ error of the standard ITE state.

Key Contributions

Quadratic Error Suppression: The paper proves that TITE quadratically suppresses the trace distance to the ground state. By replacing roughly half of the imaginary time ( $\beta \to \beta/2$ ) with randomized real-time evolution, the algorithm maintains the same level of accuracy for arbitrary observables.
Cost Reduction: Since the sample complexity and classical cost of ITE scale as $e^{O(\beta)}$ , halving the required imaginary time results in an asymptotic quadratic reduction in the total cost of ground state preparation.
Implementation Agnosticism: The TITE framework is formulated at the level of a quantum channel, making it compatible with existing ITE implementations, including Trotterization, Quantum Signal Processing (QSP), and Quantum Imaginary-Time Evolution (QITE). It does not require modifying the underlying non-unitary simulation method.
Optimal Distributions: The authors derive specific probability distributions for the random time $t$ $t$ that satisfy the necessary error suppression conditions.
- If the spectral gap lower bound $\Delta$ is known, a distribution based on the $\text{sinc}^4$ function ( $S_\Delta$ ) is optimal.
- If $\Delta$ is unknown, a Cauchy-like distribution ( $C_\beta$ ) is provided, which depends only on the chosen imaginary time $\beta$ .

Results
The authors demonstrate the efficiency of TITE through noisy circuit-level simulations of a non-integrable Ising chain. The results show substantial improvements over standard ITE, confirming that the randomized real-time evolution effectively suppresses the error in arbitrary observables without incurring the exponential overhead associated with increasing $\beta$ . The simulations validate that the algorithm achieves the predicted quadratic reduction in error relative to the trace distance.

Significance and Claims
The paper claims that TITE offers a general method to alleviate the exponential cost bottleneck of ground state preparation algorithms. By leveraging the unitary nature of real-time evolution (which incurs no sample complexity overhead) and the error-suppressing properties of randomized compiling, TITE allows for a "trade-off" where imaginary time is exchanged for randomness.

The authors position TITE as a bridge between error mitigation (suppressing physical noise) and ground state distillation (extracting pure states from noisy approximations). They emphasize that this approach mitigates algorithmic error arising from finite imaginary time without significantly increasing the computational cost relative to standard ITE. The work establishes that for a broad class of ground state preparation problems, the cost can be reduced asymptotically by a quadratic factor, making high-precision ground state estimation more feasible for near-term and early fault-tolerant quantum devices.

More like this