Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution

arXiv:2610.01954 · quant-ph, physics.chem-ph · Submitted 2026-10-01 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution".

Mira: Detailed Research Synthesis: Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution (VIME) This paper introduces Variational Imaginary-time Majorana Evolution (VIME),

Kai: First, who's behind it and why it matters.

Paper summary: Kai: So we're looking at this paper called "Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution". Essentially, they’re building a way to use classical pre-training to get the ground state of molecular electronic structures on quantum computers.

Mira: That means they took a complex problem and tried to make it easier for the actual quantum hardware you're hoping to use. The big claim here is using a single compiled Majorana-propagation surrogate graph instead of doing separate propagations for every single observable in the molecular pool.

Lev: From an error correction side, that kind of efficiency is pretty crucial because we don't have infinite qubits, so minimizing the gate count and circuit depth matters a lot when you think about running this on real hardware <ref:2610.01954#pg1>.

Kai: Right, so they are combining operator-projected variational quantum imaginary-time evolution, or OVQITE, with Majorana propagation. They’re using something called VIME to do this.

Mira: And the key innovation they focus on is pairing that VIME algorithm with a new ansatz they developed called the compressed tiled Unitary Product State, or c-tUPS <ref:2610.01954#pg1>.

Kai: What does that compression actually do? Because I remember reading about tUPS before; what makes this version better for speed and energy?

Mira: They’ve made the c-tUPS ansatz smaller by using only half as many length-four Majorana rotations, which cuts down the cost of those Majorana propagation evaluations while also giving them lower variational energies than the standard tUPS <ref:2610.01954#pg1>.

Lev: That reduction in rotation count is something we've seen before; it’s a trade-off between circuit complexity and the quality of the state they are trying to find <ref:2610.01954#pg2>.

Kai: So, what’s the actual mechanism they use to evaluate all those different things needed for the VIME update in one go? That single compiled surrogate graph you mentioned?

Mira: They are evaluating the pool expectation values, anticommutator estimates, and that surrogate Jacobian all in a single forward pass <ref:2610.01954#pg2>. This is what eliminates the overhead of running separate propagations for each observable <ref:2610.01954#pg2>.

Lev: If you're thinking about running this on a noisy machine, having that entire update happen in one go might actually help with controlling the error accumulation during the process <ref:2610.01954#pg3>.

Kai: They benchmarked this against things like ADAPT-VMPE and direct MP energy minimization, and they found pretty good results on challenging systems <ref:2610.01954#pg3>.

Mira: Across systems like the ruthenium complex TLD-one thousand four hundred eleven which needed fifty-two qubits, VIME matches or even improves upon the energy accuracy reported by ADAPT-VMPE <ref:2610.01954#pg3>.

Kai: But what about the resource costs? I want to know how much smaller this actually makes things for people building these circuits.

Mira: For the acene series, they showed comparable energy accuracy with up to twenty-six times fewer CNOTs and one hundred sixty-seven times lower two-qubit depth when compared to comparison points for ADAPT-VMPE <ref:2610.01954#pg3>.

Paper summary: Lev: Sixteen hundred seventy times lower two-qubit depth is a massive reduction, it really speaks to the hardware compatibility they are aiming for <ref:2610.01954#pg3>.

Kai: And when you look at the ground state quality itself, did this new method actually find better states than just running direct MP energy minimization?

Mira: Yes, they demonstrated that using VIME with the c-tUPS ansatz yields higher ground-state overlaps than direct MP energy minimization when using the same circuit structures <ref:2610.01954#pg3>.

Lev: Higher overlap means the state you're starting from is closer to what you actually want, which simplifies the rest of your error correction work <ref:2610.01954#pg3>.

Kai: So, if we boil this down for someone just listening to the show, what’s the actual significance of VIME and c-tUPS? What does it change for quantum chemistry or materials science right now?

Mira: It means we have a classical method that can train these variational circuits effectively enough to prepare the ground state of pretty big molecular systems, up to fifty or sixty qubits <ref:2610.01954#pg1>.

Lev: That's a significant step because preparing those states accurately is often the hardest part before you even start worrying about full error correction <ref:2610.01954#pg3>.

Kai: The authors also mentioned how they construct these four-qubit tiles, not by just picking them greedily, but through a bounded bidirectional search with a meet-in-the-middle strategy <ref:2610.01954#pg2>.

Mira: They are explicitly defining the resource bounds for different connectivity types like all-to-all or linear nearest neighbor, and they even show specific bounds for c-tUPS at four qubits, which is a lot of detail <ref:2610.01954#pg3>.

Lev: Having those explicit resource bounds is what makes it practical; you can actually plan your hardware layout before you start simulating the chemistry <ref:2610.01954#pg3>.

Kai: And there’s this adaptive extension they introduced, ADAPT-VIME, which incrementally adds Majorana rotations based on minimizing residual error <ref:2610.01954#pg3>. How does that fit into the overall picture?

Mira: It suggests a way to dynamically refine the circuit by focusing on the parts that still have the biggest error left in them, selecting the rotation with the largest positive change in energy <ref:2610.01954#pg3>.

Lev: That adaptive part is interesting because it means you aren't just running a fixed sequence; you’re optimizing for precision at every step <ref:2610.01954#pg3>.

Kai: So, to wrap this up, VIME and c-tUPS are showing that we can get high-fidelity ground state preparation for complex molecules on systems up to sixty qubits using a method trained classically <ref:2610.01954#pg1>.

Mira: It’s about making the process hardware-efficient, reducing the CNOT count by a factor of twenty-six in some cases, and improving state overlap compared to other variational methods <ref:2610.01954#pg3>.

Lev: The implication for real quantum hardware is that this method gives us a viable path toward preparing states large enough to actually test fault-tolerant algorithms <ref:2610.01954#pg3>.

Conclusion: Kai: So we’re wrapping up on this paper about Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution. The authors are showing how to get high fidelity for big molecular systems on smaller quantum hardware than before.

Mira: They’re basically proving that you can train these variational circuits classically and then use them to find the lowest energy state of really tough molecules, like those ruthenium complexes mentioned in the paper.

Lev: And from a practical standpoint, they’re showing circuit depths that are actually manageable for the near future, which is what matters when you’re trying to actually run this on real hardware.

Kai: So what does this mean for someone just listening? It means we’re getting closer to being able to use quantum computers for chemistry problems that are currently too big or too complicated.

Mira: It shifts the focus from just making the algorithm work in theory to actually building circuits that fit on the hardware you have, which is a huge hurdle right now.

Lev: The resource reduction they’re showing—like cutting CNOT counts down by factors of twenty—that’s what makes this method appealing for error correction schemes.

Kai: Exactly. So it seems like the main goal here is making the preparation step itself more efficient so we can eventually tackle those huge problems we’ve been looking at.

Mira: And they introduced this compressed ansatz, c-tUPS, which is a clever way to make the circuits smaller while keeping the energy accuracy high.

Lev: That compression helps keep the required quantum resources low enough that it doesn't immediately crash on current noisy devices.

Kai: It really does. So we’ve seen how they built these tiles and how VIME helps them train those states, but what’s next for this approach?

Federico Santona, Manuel G. Algaba, Aeishah Ameera Anuar, Anna M. Wernbacher, Prachi Sharma, Fedor Simkovic IV

IQM Quantum Computers GmbH

quant-ph, physics.chem-ph

Submitted: 2026-10-01

Updated: 2026-10-01

Comments: 52 pages, 28 figures, 16 tables

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 90/100

The gist: Detailed Research Synthesis: Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution (VIME) This paper introduces Variational Imaginary-time Majorana Evolution

Key concepts

Variational Imaginary-Time Majorana Evolution (VIME)
VIME is a classical pre-training method that uses imaginary-time evolution combined with Majorana propagation to find the ground state of quantum systems. Its main innovation is using one compiled graph to evaluate all necessary quantities for the update step, saving significant time and resources compared to older methods.
Majorana Propagation (MP)
Majorana Propagation is a technique used in this method to efficiently calculate necessary quantities like expectation values and Jacobians during the imaginary-time evolution. It allows the algorithm to perform updates based on these calculations without needing separate, costly propagations for every single observable.
c-tUPS Ansatz
The compressed tiled Unitary Product State (c-tUPS) is a specific quantum circuit structure used in this work. It is a more efficient version of the standard ansatz that uses half as many length-four Majorana rotations, which reduces the computational cost while still achieving good variational energies.
ADAPT-VMPE
ADAPT-VMPE is an established method used for benchmarking against VIME. It involves using variational Majorana propagation to find molecular ground states. The paper shows that VIME achieves comparable energy accuracy with much lower resource requirements, such as fewer CNOTs and shallower circuits.

Terminology

Summary

Detailed Research Synthesis: Hardware-Efficient Ground-State Preparation using Variational Imaginary-Time Majorana Evolution (VIME)

This paper introduces Variational Imaginary-time Majorana Evolution (VIME), a novel classical pre-training algorithm designed to efficiently prepare the ground state of molecular electronic structures using variational quantum circuits. VIME extends the framework of operator-projected variational quantum imaginary-time evolution (OVQITE) by integrating it with Majorana Propagation (MP), utilizing a single compiled surrogate graph for all necessary evaluations.

Core Methodology and Innovation

The central innovation lies in the synergy between VIME and a specialized ansatz: the compressed tiled Unitary Product State (c-tUPS).

  1. VIME Mechanism: VIME operates by combining OVQITE with MP. Crucially, it employs a single compiled surrogate graph to evaluate all quantities required for the OVQITE update—specifically, pool expectation values, anticommutator estimates, and the surrogate Jacobian (J MP)—in a joint forward pass. This eliminates the computational overhead of performing separate propagations for every observable in the molecular pool, which is standard in many variational methods.

  2. c-tUPS Ansatz: The c-tUPS ansatz is a compressed variant of the tiled Unitary Product State (tUPS). It retains the structural advantages of tUPS but achieves significant cost reduction by using only half as many length-four Majorana rotations, thereby reducing the truncation cost associated with MP evaluations while simultaneously yielding lower variational energies compared to the standard tUPS.

  3. Optimization Strategy: The VIME update is performed by simultaneously evaluating the imaginary-time velocity vector and the Jacobian (J) using this single Hamiltonian propagation graph, which is a key efficiency gain over methods requiring separate evaluations for many observables.

Comparative Performance and Benchmarking

The research rigorously benchmarks VIME against established methods, primarily ADAPT-VMPE (which uses variational Majorana-propagation) and direct MP energy minimization (VMPE).

  • Energy Accuracy: Across challenging systems—including the strongly correlated ruthenium complex TLD-1411 (requiring up to 52 qubits) and the strongly multireference acene series up to heptacene (60 qubits)—VIME consistently matches or improves upon the MP-reported energy accuracy of ADAPT-VMPE.

  • Resource Efficiency: The most compelling results demonstrate superior resource scaling. For the acene series, VIME achieves comparable energy accuracy with up to 26× fewer CNOTs and 167× lower two-qubit depth compared to ADAPT-VMPE comparison points. For TLD-1411 (52 qubits), VIME requires more than an order of magnitude fewer CNOTs and over two orders of magnitude lower two-qubit depth than ADAPT-VMPE.

  • Ground State Quality: Beyond energy, the paper demonstrates that the combination of VIME and c-tUPS yields higher ground-state overlaps than direct MP energy minimization (VMPE) using identical circuit structures.

Circuit Construction and Computational Cost Analysis

The paper provides a detailed look into the underlying quantum circuit construction methods:

  • Tile Construction: The four-qubit tiles are constructed not via greedy methods but through a bounded bidirectional search, utilizing a meet-in-the-middle strategy to build partial circuits from both ends of the rotation sequence.

  • Resource Bounds: Explicit resource bounds (NCNOT, D2Q, N1Q) are reported for different connectivity types (all-to-all, square-grid, linear nearest-neighbor) and ansatz types (c-tUPS vs. tUPS). For instance, the explicit circuits establish specific bounds for c-tUPS at 4 qubits (NCNOT=10, D2Q=8, gamma= pi/4).

  • VIME Overhead: The computational cost of VIME scales significantly with system size and circuit depth. At 64 qubits, evaluating the ITE quantities and the full pool Jacobian takes approximately 40 minutes for one layer and 2.4 hours for one-and-a-half layers, excluding the time needed to solve the linear system for parameter updates.

  • ADAPT-VIME: An adaptive extension, ADAPT-VIME, is introduced which incrementally adds Majorana rotations based on residual error minimization (1 over 2 squared), selecting the rotation with the largest positive L(lambda) a.

Conclusion and Significance

The paper concludes that VIME represents a significant advancement: it is the first demonstration of classically trained variational quantum circuits capable of achieving high-fidelity ground-state preparation for challenging molecular electronic structure problems up to 50–60 qubits, while maintaining circuit depths compatible with prospective near-term quantum hardware. By leveraging classical pre-training (VIME) and optimized circuit design (c-tUPS), the method achieves superior energy accuracy and ground state fidelity with substantially reduced quantum resource requirements compared to existing variational approaches like ADAPT-VMPE.

Improvements for AI systems

  1. The AI system can perform ground-state preparation for strongly correlated molecular electronic structures up to 60 qubits with chemical precision against DMRG reference energies, achieving chemical precision for systems like TLD-1433 (52 qubits) and acenes (60 qubits).

  2. The system can achieve significant circuit resource reduction compared to existing methods, specifically utilizing VIME with c-tUPS to reduce two-qubit depth by factors of 36–167 and CNOT counts by factors of 3.3–26 for the acene series up to heptacene (60 qubits).

  3. The AI system can operate efficiently in a classical pre-training regime, as VIME matches or improves on the MP-reported energy accuracy of the ADAPT-VMPE comparison points with up to 26× fewer CNOTs and 167× lower two-qubit depth.

  4. The system can leverage a single compiled Hamiltonian surrogate graph to evaluate all quantities required for the Variational Imaginary-time Majorana Evolution (VIME) update, which eliminates the need to perform a separate propagation for each operator in the molecular pool.

  5. The AI system can implement adaptive imaginary-time integration using Armijo backtracking, allowing it to select an imaginary-time step size based on whether a sufficiently small step along ˙θ lowers the surrogate energy by checking if "sE < −ϵE."

Abstract

Compact ground-state preparation circuits are essential for early fault-tolerant quantum chemistry. We introduce Variational Imaginary-time Majorana Evolution (VIME), a classical pre-training algorithm that extends operator-projected variational quantum imaginary-time evolution (OVQITE) to molecular electronic structure using a single compiled Majorana-propagation (MP) surrogate graph. We pair VIME with a compressed tiled Unitary Product State (c-tUPS) ansatz that we develop to reduce both classical simulation and quantum state-preparation costs. In the strongly correlated ruthenium complex TLD-1411, active-space calculations requiring up to 52 qubits achieve chemical precision against DMRG reference energies, with energy errors approximately 400 times smaller than those from ADAPT-VQE-based variational Majorana-propagation (ADAPT-VMPE). Across the strongly multireference acene series up to heptacene (60 qubits), VIME matches or improves on the MP-reported energy accuracy of the ADAPT-VMPE comparison points with up to 26 times fewer CNOTs and 167 times lower two-qubit depth. These results establish imaginary-time classical pre-training of compressed variational ansätze as a promising route to resource-efficient molecular ground-state preparation.

Sources

Related papers