Repository navigation
TensorCircuit-NG v1.10.0
This release adds new tensor-network algorithms and expands TensorCircuit-NG’s matrix-function and spectral tools, alongside a broad set of correctness and stability improvements.
Highlights
-
Added
tensorcircuit.tnalgexperimental module with fixed-shape JAX implementations of MPS TEBD, one-site TDVP, and one-site DMRG. It supports dense and Abelian block-buffer PyTree states, Pauli-string MPO builders, checkpoints, and interoperability with legacy MPS states. -
Expanded matrix-function and spectral methods for Krylov/SLQ, Chebyshev/KPM, and Taylor calculations, including reusable JIT- and AD-compatible queries and density-of-states, thermal, and response APIs.
-
Unified the public time-evolution entry points with the matrix-function implementation.
Fixes and improvements
-
Corrected complex-gradient handling across JAX, PyTorch, and TensorFlow, and improved DLPack transfers for noncontiguous PyTorch tensors.
-
Fixed circuit behavior involving AnalogCircuit initial states and composition, periodic Line1D weights, QAOA Ising cost angles, and time-dependent analog inversion.
-
Improved measurement, observable, and noise semantics across ZX and stabilizer circuits, including Stim imports, MPP readout noise, and sampled Pauli expectations.
-
And many more stability improvements and bug fixes...
Acknowledgments
Special thanks to @QingyunQian for identifying and fixing many bugs throughout this release.