The Signal
Preprint tests disorder-guided circuit on 49-qubit IQM model
Researchers ran a disorder-guided variational circuit on an IQM processor for a 49-qubit model and report 97.09% relative energy accuracy. Unlike their smaller-system trial, the 49-qubit run used parameters optimized in advance on classical computers.

A 49-qubit test, with a classical step
A 49-qubit variational circuit was run on an IQM processor to estimate the energy of a two-dimensional disordered Ising model. In a preprint submitted to arXiv on November 16, Alina Joch and Benedikt Fauseweh report 97.09% ± 0.07% relative energy accuracy for that hardware run.
The result used parameters optimized in advance with classical simulations; the processor executed the resulting circuit and measured its energy. The authors distinguish this from their 16-qubit test, where energy and gradients were measured on the processor during five rounds of optimization.
Using renormalization to shape the circuit
The study concerns the random transverse-field Ising model on a two-dimensional square lattice. Its couplings and transverse fields are drawn independently from a disorder distribution. The researchers selected model instances near the median pseudocritical point for the relevant system size.
The method adapts the emergent-coupling-based ansatz, or ECBA, for this model. Strong-disorder renormalization identifies dominant couplings and builds an approximate low-energy state; the resulting structure guides the circuit, while adjustable gates let the state move beyond that starting approximation. The circuit includes entangling operations between selected qubit pairs and rotations on individual qubits.
In classical simulations, the paper compares ECBA with the renormalization-group starting state and a nearest-neighbor ansatz. The authors report that ECBA produced lower energy errors than both alternatives in the cases they studied. For the tested configurations, its circuit depth averaged 21 ± 1, while the nearest-neighbor circuit averaged 32 ± 5.
Two device runs, different workflows
The hardware experiments used IQM’s 20-qubit Garnet and 54-qubit Emerald superconducting processors. The 16-qubit experiment ran on Garnet and reached 89.38% ± 2.26% relative energy accuracy after five on-device optimization iterations. The paper says this was close to the 87.98% accuracy in a simulation limited to the same number of iterations.
For the 49-qubit experiment, the team used Emerald and classically pre-optimized parameters. That approach left most of the optimization computation to classical simulation, rather than evaluating the full optimization loop on the processor. The result therefore demonstrates hardware execution of the larger circuit with pre-optimized settings, not a 49-qubit on-device optimization run.
Correlations the circuit misses
The paper also compares measured correlation patterns with tensor-network calculations. For the 49-qubit system, the researchers say the hardware results reproduce correlations around renormalization-group clusters, but correlations between some unconnected groups are absent. They cite missing correlations between spin 7 and spins 42 and 43 as examples.
The authors say capturing those additional correlations would require adjusting circuit-selection parameters, which could add variational parameters and increase circuit depth. They present the work as a test of applying renormalization-guided variational circuits to a two-dimensional disordered model, while noting that the larger hardware run relied on classical pre-optimization.
Sources
Related stories
- Four Delft researchers win ERC grant for spin-qubit project
- Quantum team tests monopole-scattering analogue on Quantinuum H2
- Chalmers Team Uses Parasitic Capacitances to Cut Qubit Crosstalk
- Researchers map energy levels in a silicon quantum-dot pair
- IBM says revised code and decoder advance its path to Starling