Expand description
Fixed-rank Gaussian range finding and dense randomized SVD.
Composes the backend’s contractions, QR and SVD following Halko, Martinsson and Tropp, SIAM Review 53 (2011), Algorithms 4.4 and 5.1. No upstream source code was translated. The input is a materialized local tensor; these routines do not materialize a tensor network.
Only dense f64 and Complex64 inputs are supported. f32 and Complex32
(c32) return FactorizeError::UnsupportedStorage before consuming RNG state.
Structs§
- Range
Options - Sketch size and numerical stabilization for a Gaussian range finder.
- Rsvd
Options - Fixed output rank and sketch options for
randomized_svd. - Rsvd
Result - Approximate SVD factors with directly contractible index identities.
- Rsvd
Timings - Time spent in one RSVD composite, including sketch construction and index work.
Functions§
- randomized_
range - Find an orthonormal basis for a dense tensor’s left-index range.
- randomized_
range_ in - Find a Gaussian range basis in a caller-owned execution context.
- randomized_
range_ on_ session - Find a Gaussian range basis on a caller-held plain session.
- randomized_
svd - Compute a fixed-rank randomized SVD of a materialized local tensor.