Speaker
Description
Quantics Tensor Cross Interpolation (QTCI) provides an efficient tensor-network representation of high-resolution functions by encoding physical coordinates into binary degrees of freedom and constructing a tensor train directly from sampled function values. Although QTCI is effective for many structured functions, its efficiency can deteriorate when the target function develops strong correlations across the binary variables, leading to large tensor-train bond dimensions. To address this limitation, we introduce a disentangling scheme based on invertible bit-string transformations generated from a GL(2,2)-based binary gate set. By reorganizing the binary coordinates before interpolation, this procedure seeks a representation in which the same sampled function admits a more disentangled tensor-train structure. We show that this disentangling approach improves the compression of otherwise high-rank functions at comparable accuracy, reducing the required bond dimensions and extending the practical applicability of QTCI to more challenging high-resolution functions.