Imagine you are trying to find the lowest point in a vast, foggy mountain range. In the world of physics, this "lowest point" is called the ground state, and the higher peaks are excited states. Knowing where these points are helps scientists understand how materials behave, how magnets work, and how quantum computers function.

For a long time, finding these points on a computer has been like trying to map every single inch of that mountain range with a ruler. As the mountain gets bigger (more particles involved), the task becomes impossible for classical computers because the amount of data explodes.

This paper introduces a new, "penalty-free" quantum algorithm that acts like a smart, automated drone to find these points. Here is how it works, broken down into simple concepts:

1. The Problem with Old Methods

Most current methods are like trying to find the lowest point by guessing and checking. You build a model, guess a location, and then use a classical computer to tweak your guess.

The Trap: Sometimes, the computer gets stuck in a "barren plateau"—a flat area where no matter which way you nudge your guess, it doesn't get better. It's like walking on a flat desert and not knowing which direction leads to the valley.
The Penalty: To find the second lowest point (the first excited state), old methods often have to add a "penalty" to the math. It's like putting a giant boulder on the lowest point so the drone is forced to ignore it and look for the next one. This boulder is hard to build and often breaks the system.

2. The New Approach: The "Stochastic Sampler"

The authors propose a method that doesn't guess, doesn't use penalties, and doesn't need a classical computer to help. It relies on Imaginary Time Evolution (ITE).

Think of ITE as a magical filter that slowly drains the "energy" out of a system. If you start with a random mix of states, this filter naturally drains away the high-energy states, leaving only the lowest energy state behind.

How they make it work on a quantum computer:
Instead of trying to build a giant, complex machine to drain the energy all at once, they break the problem into two smaller, easier pieces (let's call them Piece A and Piece B).

Imagine you have a complex puzzle, but you know the solution to the left half and the solution to the right half separately.
The algorithm randomly picks a piece (either A or B) and applies a tiny bit of "draining" to it.
By repeating this random sampling thousands of times, the system naturally flows toward the ground state. It's like a drop of water rolling down a hill; it doesn't need a map, it just follows the path of least resistance.

3. Finding the Higher Peaks (Excited States)

Once the drone finds the lowest valley (the ground state), how do we find the next lowest one without using a "penalty boulder"?

The authors use a clever trick called State-Based Simulation.

The Analogy: Imagine you have found the lowest valley. Now you want to find the second lowest. Instead of putting a boulder on the first valley, you make a perfect "ghost copy" of that valley and place it next to the real one.
The algorithm then performs a special dance (a quantum operation) between the real system and this ghost copy. If the real system looks too much like the ghost copy (the ground state), the dance cancels it out.
This effectively "filters out" the ground state, allowing the system to naturally settle into the next lowest valley (the first excited state).
You can repeat this process: once you find the second valley, you make a ghost copy of it, filter it out, and find the third.

4. Why This is a Big Deal

No Penalties: It doesn't need to add artificial "boulders" (penalty functions) to force the system to ignore the ground state. It just filters them out cleanly.
No Barren Plateaus: Because it doesn't rely on a classical computer to tweak parameters (like the old "guess and check" methods), it avoids the trap of getting stuck in flat, unhelpful areas.
Purely Quantum: It runs entirely on the quantum computer, using the natural properties of quantum mechanics to do the heavy lifting.

5. The Proof

The authors tested this idea using a famous model called the Transverse Ising Model (think of it as a row of tiny magnets that can flip up or down).

They successfully found the ground state and the first three excited states.
The results were very accurate (over 96% fidelity), even when they simulated a larger system with 10 magnets.
They showed that even if the magnets are nearly identical in energy (nearly degenerate), the algorithm can still tell them apart.

Summary

This paper presents a new way to use a quantum computer to solve complex energy problems. Instead of struggling with penalties and getting stuck in dead ends, this method uses random sampling to flow naturally to the lowest energy state, and then uses ghost copies to filter out what it has already found, revealing the next level of energy. It's a cleaner, more direct path to understanding the quantum world.

Technical Summary: A Penalty-Free Quantum Algorithm to Find Energy Eigenstates

Problem Statement

Finding the eigenvalues and eigenstates of many-body Hamiltonians is a fundamental challenge in physics and computational science. Classical methods face exponential scaling ( $O(2^{2n})$ storage and $O(2^{3n})$ time) as the number of qubits $n$ increases. While quantum algorithms offer a path forward, existing approaches have significant limitations:

Variational Quantum Eigensolver (VQE): A hybrid quantum-classical approach that often suffers from "barren plateaus" (vanishing gradients) during optimization. Finding excited states typically requires penalty terms to suppress ground-state contributions, increasing complexity and potential for error.
Quantum Annealing: Requires long evolution times to ensure adiabaticity and struggles with small energy gaps between the ground and excited states.
Imaginary Time Evolution (ITE) & QITE: While promising, standard ITE faces scalability issues in realizing non-unitary operators. Quantum ITE (QITE) often relies on variational optimization or penalty terms for excited states, reintroducing the challenges of barren plateaus and requiring complex ansatzes.
Other Methods: Techniques like Quantum Filter Diagonalization (QFD) and Filtered Quantum Phase Estimation (FQPE) depend heavily on specific reference states or optimized filter functions.

The authors identify a need for a fully quantum algorithm that can find both ground and excited states without penalty functions, variational steps, or hybrid classical-quantum loops, thereby avoiding barren plateaus.

Methodology

The proposed algorithm is a fully quantum, penalty-free method based on stochastic sampling of a mixed-state density matrix combined with state-based simulation.

1. Hamiltonian Decomposition and Density Matrix Mapping

The algorithm assumes the target Hamiltonian $H$ can be decomposed into two solvable components, $H = h_1 + h_2$ , where the eigenvalues ( $\lambda^j_i$ ) and eigenstates ( $|\psi^j_i\rangle$ ) of $h_1$ and $h_2$ are known and easy to prepare.

The authors shift and rescale the eigenvalues of $h_1$ and $h_2$ to create a valid probability distribution.
They construct a mixed-state density matrix $\rho$ equivalent to the shifted Hamiltonian:
$\rho = \frac{1}{C} (H + (c_1 + c_2)I) = \sum_{i,j} p^j_i |\psi^j_i\rangle\langle\psi^j_i|$
where $p^j_i$ are probabilities derived from the shifted eigenvalues. The eigenstates of $\rho$ are identical to those of $H$ .

2. Stochastic Imaginary Time Evolution (ITE) for the Ground State

To find the ground state, the algorithm performs ITE on $\rho$ using stochastic sampling:

Sampling: At each time step $\eta$ , a projector $P^j_i = |\psi^j_i\rangle\langle\psi^j_i|$ is sampled from the distribution $\{p^j_i\}$ .
Evolution Operator: Instead of implementing the full non-unitary $e^{-\eta \rho}$ , the algorithm implements the operator $I - \eta P^j_i$ for the sampled projector.
Circuit Implementation: This is achieved via a post-selected unitary circuit involving an ancillary qubit. The unitary $U$ is an $n$ -qubit controlled-SU(2) gate. If the system is in the state corresponding to the sampled projector, the ancilla undergoes a rotation; otherwise, it remains unchanged. Post-selecting the ancilla in the $|0\rangle$ state yields the desired $I - \eta P^j_i$ operation.
Convergence: Repeating this process drives the system toward the ground state of $\rho$ (and thus $H$ ) without variational optimization.

3. State-Based Simulation for Excited States

To find excited states, the algorithm must suppress the contributions of already-found lower-energy states (e.g., the ground state $P_g$ ).

Challenge: Directly implementing projectors for highly entangled ground states is difficult.
Solution: The authors utilize a state-based simulation protocol. This involves a dual $n$ -qubit system (ancilla) prepared in the known ground state $P_g$ , a control qubit, and the primary system.
Mechanism: A controlled-SWAP gate is applied between the primary system and the ancilla, conditioned on the control qubit. Measuring the control qubit in the $|+\rangle$ basis effectively implements the evolution $e^{-\eta P_g}$ (approximated as $I - \eta P_g$ ), projecting out the ground state component.
Lifting Procedure: To find the first excited state, the probability of sampling the ground state projector is artificially "lifted" (increased) in the density matrix distribution. The algorithm then performs ITE on this modified distribution. When the ground state projector is sampled, the state-based simulation is used to filter it out. This process can be iterated to find higher excited states.

4. Variant for General Hamiltonians

For Hamiltonians not easily split into two solvable parts, the authors propose a variant where $H$ is treated as a linear combination of Pauli operators ( $H = \sum \alpha_i O_i$ ). Stochastic sampling is applied directly to these Pauli operators, with similar circuit implementations for the evolution steps.

Key Results

The authors validated the algorithm using classical simulations of the Transverse Field Ising Model (a 1D chain with periodic boundary conditions), which serves as a benchmark because it can be split into solvable $Z$ -basis and $X$ -basis components.

Ground State (4-qubit system): Achieved a fidelity of 99.9% with the exact ground state.
Excited States (4-qubit system):
- 1st excited state: 98.6% fidelity.
- 2nd excited state: 96.8% fidelity.
- 3rd excited state: 96.6% fidelity.
Scalability Test (10-qubit system): The algorithm successfully found the ground state (98.6% fidelity) and the first excited state (97.8% fidelity). It also resolved the third lowest eigenstate (90.2% fidelity) despite near-degeneracy between the first two states.
Pauli Variant: The Pauli-operator sampling variant also demonstrated high fidelities (97.1%–99.5%) for the first four eigenstates in the 4-qubit case.
Resource Scaling: Theoretical analysis indicates the algorithm scales with $O(n)$ two-qubit gates per step, a significant improvement over Trotter-based ITE methods which scale as $O(n^2)$ .

Significance and Claims

The paper claims to present a fully quantum algorithm that addresses the eigenstate problem without the drawbacks of current hybrid or variational methods:

Penalty-Free: It finds excited states without adding penalty terms to a loss function, avoiding the need for complex ansatzes and the risk of barren plateaus.
No Hybrid Steps: It eliminates the need for classical optimization loops, making it a purely quantum procedure.
Broad Applicability: The assumption that $H$ can be split into solvable parts is argued to be non-restrictive, covering many physical models and even QMA-complete problems.
Robustness: The method demonstrates the ability to find multiple low-lying eigenstates sequentially, even in systems with near-degenerate energy levels.

The authors acknowledge that the overall success rate depends on post-selection, which may be a challenge on current Noisy Intermediate-Scale Quantum (NISQ) devices due to gate errors and decoherence. However, they posit that the method offers a concrete pathway for using quantum circuits to tackle classically intractable eigenstate problems, serving as a foundational addition to the quantum computational toolbox.

A penalty-free quantum algorithm to find energy eigenstates