Struct TensorCI2 

pub struct TensorCI2<T: Scalar + TTScalar> { /* private fields */ }

State object for the two-site Tensor Cross Interpolation algorithm.

Holds the current index sets (I, J), site tensors, and error statistics produced by crossinterpolate2. After interpolation, call to_tensor_train to extract the resulting TensorTrain.

Key methods

Method	Purpose
rank	Maximum bond dimension
link_dims	Bond dimensions at each bond
to_tensor_train	Extract the tensor train
max_bond_error	Largest bond error from last sweep
pivot_errors	Per-bond pivot errors from back-truncation

You normally do not construct TensorCI2 directly; use crossinterpolate2 instead.

Implementations

impl<T> TensorCI2<T>

where T: Scalar + TTScalar + Default + MatrixLuciScalar, DenseFaerLuKernel: PivotKernel<T>, LazyBlockRookKernel: PivotKernel<T>,

pub fn new(local_dims: Vec<usize>) -> Result<Self>

Create a new empty TensorCI2

pub fn len(&self) -> usize

Number of sites

pub fn is_empty(&self) -> bool

Check if empty

pub fn local_dims(&self) -> &[usize]

Get local dimensions

pub fn rank(&self) -> usize

Get current rank (maximum bond dimension)

pub fn link_dims(&self) -> Vec<usize>

Get bond dimensions

pub fn max_sample_value(&self) -> f64

Get the maximum observed sample value seen so far

pub fn max_bond_error(&self) -> f64

Get maximum bond error

pub fn pivot_errors(&self) -> &[f64]

Get pivot errors from back-truncation

pub fn is_site_tensors_available(&self) -> bool

Check if site tensors are available

pub fn site_tensor(&self, p: usize) -> &Tensor3<T>

Get site tensor at position p

pub fn to_tensor_train(&self) -> Result<TensorTrain<T>>

Convert to TensorTrain

pub fn add_global_pivots(&mut self, pivots: &[MultiIndex]) -> Result<()>

Add global pivots to the TCI

pub fn i_set(&self, p: usize) -> &[MultiIndex]

Get I-set at site p

pub fn j_set(&self, p: usize) -> &[MultiIndex]

Get J-set at site p

pub fn invalidate_site_tensors(&mut self)

Invalidate all site tensors

pub fn flush_pivot_errors(&mut self)

Flush pivot errors (reset to empty)

pub fn sweep2site<F, B>( &mut self, f: &F, batched_f: &Option<B>, forward: bool, options: &TCI2Options, ) -> Result<()>

where F: Fn(&MultiIndex) -> T, B: Fn(&[MultiIndex]) -> Vec<T>,

Perform one 2-site sweep.

This is a public wrapper around the internal update_pivots logic, suitable for calling from C-API.

pub fn sweep1site<F>( &mut self, f: &F, forward: bool, rel_tol: f64, abs_tol: f64, max_bond_dim: usize, update_tensors: bool, ) -> Result<()>

where F: Fn(&MultiIndex) -> T,

Perform a 1-site sweep, updating I/J sets and optionally site tensors.

This is used for cleanup after adding global pivots, and for computing canonical site tensors.

Port of Julia’s sweep1site! from tensorci2.jl.

pub fn fill_site_tensors<F>(&mut self, f: &F) -> Result<()>

where F: Fn(&MultiIndex) -> T,

Fill all site tensors using 1-site LU decomposition at each bond.

For each site b (except the last), computes the Pi matrix (Kronecker(I_b, d_b) × J_b) and the pivot matrix P (I_{b+1} × J_b), then solves P^T \ Pi^T to get the site tensor T_b = Pi * P^{-1}. The last site tensor is set by direct evaluation.

Port of Julia’s fillsitetensors! / setsitetensor!.

pub fn make_canonical<F>( &mut self, f: &F, rel_tol: f64, abs_tol: f64, max_bond_dim: usize, ) -> Result<()>

where F: Fn(&MultiIndex) -> T,

Make the TCI canonical by performing 3 one-site sweeps.

1. Forward sweep (exact, no truncation)
2. Backward sweep (with truncation)
3. Forward sweep (with truncation + update tensors)

Port of Julia’s makecanonical!.

Trait Implementations

impl<T: Clone + Scalar + TTScalar> Clone for TensorCI2<T>

fn clone(&self) -> TensorCI2<T>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T: Debug + Scalar + TTScalar> Debug for TensorCI2<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.