QTT Diagnostics: Tensor Train Analysis

• Quantics Tensor Train diagnostics is the application of QTT representations to assess the multiscale structure of quantics-folded tensors using bond dimensions and singular-value spectra.
• It utilizes TT-based methodologies such as rank profiling, reconstruction scores, and error metrics to capture and quantify regular versus complex data features.
• These diagnostics enable practical applications including anomaly detection, phase transition analysis, and enhanced multiscale PDE solvers while ensuring numerical stability and efficiency.

Searching arXiv for the cited papers on QTT diagnostics, TT anomaly detection, TT-cross error analysis, and related QTT applications. Quantics Tensor Train diagnostics is the use of Quantics Tensor Train (QTT) representations together with bond dimensions, singular-value spectra, reconstruction statistics, conditioning indicators, and problem-specific residuals to assess whether a quantics-folded tensor preserves the relevant structure of data, correlators, operators, or numerical solutions. In this setting, QTT is Tensor Train (TT) applied to a folded tensor with logarithmically small modes, so TT diagnostics carry over directly: low ranks indicate simple recurring structure, inflated ranks indicate increased complexity, singular-value decay controls truncation, and application-dependent observables such as normalized inner products, entropy-like quantities, conservation laws, or cross-conditioning parameters become the operative diagnostics (Ali et al., 2024, Rohshap et al., 15 Jul 2025, Qin et al., 2022).

1. Representation, folding, and the meaning of ranks

For a $d$-way tensor $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$, the TT decomposition writes

$X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$

where each core $G_k(i_k)$ is an $r_{k-1}\times r_k$ matrix and the boundary ranks satisfy $r_0=r_d=1$. The tuple $r=(r_1,\ldots,r_{d-1})$ is the TT rank vector. Equivalent formulations use three-way cores $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$ and matricizations $X_{1:k}$, with $r_k=\mathrm{rank}(X_{1:k})$. In this language, TT ranks are the bond dimensions separating the first $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$0 modes from the remaining $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$1 modes. Low ranks encode regularity; large ranks signal stronger inter-mode correlation or structural irregularity (Ali et al., 2024, Krämer, 2017).

QTT replaces a large physical index by many small indices. For a vector $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$2 with $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$3, quantics folding maps $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$4 to an $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$5-way tensor $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$6 with

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$7

More generally, if $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$8, one reshapes into $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$9. The resulting mode sizes are small, often binary, while the number of modes grows only logarithmically with the original resolution. This is the structural basis for the frequent statement that QTT achieves exponential resolution in the original variable while keeping mode sizes fixed (Ali et al., 2024, Ishida et al., 2024).

The diagnostic meaning of TT singular values is inherited unchanged by QTT. If $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$0 denotes the $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$1-th unfolding, then the TT singular values are the singular values of $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$2, and the trace property gives

$X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$3

For QTT, these singular values are computed on the folded tensor, so the same spectra diagnose multiscale structure in the quantized coordinates rather than in the original unfurled array (Krämer, 2017).

In several recent applications, these ranks and spectra are given direct physical or statistical interpretations. In correlator compression, they quantify “time-scale entanglement”; in anomaly detection, they quantify how much unusual samples force additional singular values to survive truncation; in numerical time evolution, they track the growth of multiscale complexity and the onset of noise-dominated regimes (Rohshap et al., 15 Jul 2025, Ali et al., 2024, Ye, 13 May 2026).

2. Core diagnostic quantities

Recent work uses a relatively small set of recurrent diagnostic quantities. They differ by application, but they are all derived from the same TT/QTT objects: ranks, singular values, orthogonality relations, and compressed reconstructions.

Diagnostic	Definition	Typical interpretation
TT/QTT ranks $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$4	Bond dimensions after truncation	Structural complexity across a cut
Reconstruction score $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$5	$X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$6	Alignment preserved by compression
Relative error $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$7	$X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$8	Loss of structure under truncation
Time-scale entropy $X(i_1,\ldots,i_d)=G_1(i_1)G_2(i_2)\cdots G_d(i_d),$9	$G_k(i_k)$0	Entanglement across time partitions
Compression ratio $G_k(i_k)$1	dense storage / TT-core storage	Regularity at fixed fidelity
Conditioning $G_k(i_k)$2	Norms of pseudoinverses/interpolants	Stability of TT-cross selection
Norm/orthogonality checks	$G_k(i_k)$3, $G_k(i_k)$4	Gauge quality and numerical stability

In TT-SVD-based anomaly detection, the main decision statistic is the normalized inner product

$G_k(i_k)$5

where $G_k(i_k)$6 is the compressed-and-recovered sample. Residual-style quantities are also natural: $G_k(i_k)$7 Steep singular-value decay and stable $G_k(i_k)$8 indicate regular structure; shallow decay, rank growth, or low $G_k(i_k)$9 indicate complexity or anomaly. Compression ratio

$r_{k-1}\times r_k$0

is used as an auxiliary diagnostic: normal data often attain high $r_{k-1}\times r_k$1 without losing $r_{k-1}\times r_k$2, whereas anomalous samples lose alignment when truncation is strong (Ali et al., 2024).

In QTT diagnostics for phase transitions, the singular values $r_{k-1}\times r_k$3 of the unfolding at cut $r_{k-1}\times r_k$4 are normalized as

$r_{k-1}\times r_k$5

and the associated entropy is

$r_{k-1}\times r_k$6

Two rank-based aggregate measures are used: $r_{k-1}\times r_k$7 Peaks or sharp changes in $r_{k-1}\times r_k$8, $r_{k-1}\times r_k$9, or $r_0=r_d=1$0 quantify enhanced time-scale entanglement and are used to identify quantum phase transitions and thermal crossovers (Rohshap et al., 15 Jul 2025).

For TT-cross and QTT-cross reconstructions, diagnostics are dominated by conditioning. With compact SVD factors $r_0=r_d=1$1 and $r_0=r_d=1$2 and selected row and column sets $r_0=r_d=1$3, $r_0=r_d=1$4, the conditioning parameters

$r_0=r_d=1$5

$r_0=r_d=1$6

and the spectral parameter $r_0=r_d=1$7 diagnose whether the chosen crosses align with the dominant singular subspaces and whether the intersection matrices are well-conditioned. Large values indicate unstable interpolation, sensitivity to noise, or the need for stronger truncated pseudoinverses $r_0=r_d=1$8 (Qin et al., 2022).

In normalized TT geometry, diagnostic emphasis shifts to the unit-norm constraint and core orthogonality. The norm defect

$r_0=r_d=1$9

should vanish on the normalized TT manifold, and left-orthogonality is checked through

$r=(r_1,\ldots,r_{d-1})$0

These quantities detect gauge deterioration and ill-conditioned interface matrices during manifold optimization (Peng et al., 6 Nov 2025).

3. Construction paradigms and diagnostic workflows

A first paradigm is TT-SVD with a single global truncation factor $r=(r_1,\ldots,r_{d-1})$1. At each unfolding, singular values $r=(r_1,\ldots,r_{d-1})$2 are kept if

$r=(r_1,\ldots,r_{d-1})$3

Larger $r=(r_1,\ldots,r_{d-1})$4 produces stronger truncation and lower ranks; smaller $r=(r_1,\ldots,r_{d-1})$5 preserves more detail. In QTT this same rule acts on the quantics-folded tensor, and the resulting rank profile, singular-value decay, and reconstruction score become the diagnostic output. This is the basis of the TT/QTT anomaly-detection framework and of many compression-based workflows (Ali et al., 2024).

A second paradigm is TT-cross or QTCI, where the tensor is reconstructed from adaptively selected subtensors rather than from full unfoldings. The theoretical analysis emphasizes that global Frobenius-norm error depends polylogarithmically on the order $r=(r_1,\ldots,r_{d-1})$6 under balanced recursion, suitable rank conditions, and good conditioning of the selected crosses. Truncated pseudoinverses $r=(r_1,\ldots,r_{d-1})$7 regularize overspecified rank and noisy sampling, and the practical diagnostics are therefore local residuals, holdout errors, estimates of $r=(r_1,\ldots,r_{d-1})$8, $r=(r_1,\ldots,r_{d-1})$9, $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$0, and $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$1, and sensitivity tests such as leave-one-cross-out (Qin et al., 2022).

A third paradigm, designed explicitly for noisy evaluations, combines QTCI with nonlinear least squares. The procedure is: run QTCI to collect measured points, compress the interpolant from $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$2 to $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$3 by TT-SVD, then fit the TT cores to all measured points by minimizing a least-squares objective. In the reported implementation, automatic differentiation and LBFGS are used, and the working cost is

$G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$4

Here the diagnostics are residual norms, RMSE, reduced $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$5 when noise variances are available, pivot stability as $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$6 grows, and the separation between interpolation error and optimization error (Sakaue et al., 2024).

A fourth paradigm appears in quantics solvers tailored to multiscale PDEs and eigenproblems. Standard two-site ALS or DMRG is modified by multigrid-inspired restriction $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$7, prolongation $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$8, and a coarse-to-fine schedule that resolves long-wavelength structure before fine details. Diagnostics include residual norms $G^{(k)}\in K^{r_{k-1}\times n_k\times r_k}$9, energy error, virial ratios, parameter counts, and consistency across restriction levels. This suggests that, in QTT, diagnostics are often embedded directly into the solver architecture rather than applied only after compression (Li et al., 10 Apr 2026).

4. Data analysis, anomaly detection, and local surrogates

The most explicit diagnostic framework in data analysis is the TT-based anomaly-detection scheme that compresses data so that normal structure is preserved while anomalous structure is deleted or strongly attenuated. The paper formulates four detectors: ACGCTNAD and GCGCTNAD for global compression, ACLCTNAD and GCLCTNAD for local compression tied to a normal reference sample. All use either the auto-comparative score

$X_{1:k}$0

or the group-comparative score

$X_{1:k}$1

followed by thresholding. QTT is explicitly described as a natural extension because it is TT on quantics-folded data, often with stronger compression and lower ranks because modes are logarithmically small. Empirically, ACGCTNAD is about $X_{1:k}$2 faster than ACLCTNAD; on digits, maximum AUROC across digits is at least $X_{1:k}$3 and best $X_{1:k}$4; on the cybersecurity dataset, supervised ACGCTNAD without scaling at $X_{1:k}$5 gives AUROC $X_{1:k}$6 and accuracy $X_{1:k}$7, while standard scaling “ruins the results significantly” in that tabular setting (Ali et al., 2024).

TT-cross diagnostics also appear in quantum state tomography, where the target is a matrix product operator rather than a quantics-folded vector. The practical indicators are the sampled discrepancy

$X_{1:k}$8

the condition numbers $X_{1:k}$9 of the selected skeleton blocks, the profile of chosen ranks $r_k=\mathrm{rank}(X_{1:k})$0, and, when accessible, the normalized Frobenius error, fidelity, and trace distance. The number of measurement settings scales approximately as $r_k=\mathrm{rank}(X_{1:k})$1, so diagnostics are used both to monitor accuracy and to control measurement economy (Lidiak et al., 2022).

In local surrogates for trained quantum learning models, the diagnostic structure is explicitly decomposed into three error sources: Taylor truncation, TT approximation, and statistical estimation. The deterministic certificate is

$r_k=\mathrm{rank}(X_{1:k})$2

with

$r_k=\mathrm{rank}(X_{1:k})$3

The paper does not treat QTT directly, but it gives a theoretically motivated adaptation in which each length-$r_k=\mathrm{rank}(X_{1:k})$4 feature vector is folded into $r_k=\mathrm{rank}(X_{1:k})$5 quantized modes. The feature-norm bound $r_k=\mathrm{rank}(X_{1:k})$6 is unchanged, while the effective parameter count changes from $r_k=\mathrm{rank}(X_{1:k})$7 to approximately $r_k=\mathrm{rank}(X_{1:k})$8 (Nair et al., 28 Apr 2026).

5. Correlator compression, phase transitions, and nonequilibrium many-body dynamics

In many-body physics, QTT diagnostics are often formulated directly in terms of correlators. The method termed QTTD compresses multi-point correlators into QTTs and reads off “time-scale entanglement” from the resulting bond dimensions and singular-value spectra. The practical criterion is sharp, $r_k=\mathrm{rank}(X_{1:k})$9-stable peaks in $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$00 or $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$01 at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$02 for quantum phase transitions, and broader, temperature-dependent maxima for thermal crossovers. In the Hubbard dimer, QTT bond-dimension peaks lie precisely at singlet–doublet ground-state transition lines; in the four-site Hubbard ring, a broad QTT maximum aligns with $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$03, inflection points of negativity and mutual information, and rapid change in double occupancy; in the single-impurity Anderson model, broad maxima in $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$04 align with estimates of the Kondo temperature (Rohshap et al., 15 Jul 2025).

For diagrammatic integrands in multiorbital electron-phonon models, QTT is combined with Tensor Cross Interpolation. The diagnostics are the bond-dimension profile $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$05, the max-norm interpolation error

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$06

cross-index stability, and symmetry checks such as Nambu relations. The reported rank profiles peak near boundaries between time scales $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$07 and $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$08, then decrease toward shorter time scales. The interpolation error exhibits “faster-than-power-law” convergence; for example, at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$09, reaching $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$10 requires $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$11, whereas in the superconducting phase at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$12 it requires $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$13. Global updates are introduced because local two-site updates alone can suffer from an ergodicity problem in discrete orbital and phonon sectors (Ishida et al., 2024).

In nonequilibrium Green’s-function calculations and self-consistent $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$14, the operative diagnostics are bond dimensions $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$15, truncation tolerances, convergence measures, and physics-side conservation laws. In the $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$16 study, the per-iteration convergence precision is

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$17

with representative converged values around $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$18–$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$19. Particle number conservation is accurate to about $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$20, energy pumping is consistent to about $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$21 outside the pulse, and fermionic sum rules hold at about the $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$22 level. The same work reports that decompressing a fine-grid $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$23 would require $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$24 exabytes, while the QTT-compressed representation uses only $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$25 GB; bond dimensions saturate with $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$26 in quench calculations and are controlled by $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$27 under strong driving (Środa et al., 2024). In the earlier second-order Hubbard implementation, calculations with compressed two-time functions are reported to be possible “without any loss of accuracy,” and the QTT implementation shows much improved scaling of computational effort and memory demand with contour length (Murray et al., 2023).

6. PDEs, time integration, and multiscale scientific computing

In QTT solvers for PDEs, diagnostics are inseparable from numerical stability. In the split-step QTT solver for the time-dependent Gross–Pitaevskii equation, the kinetic step uses a low-pass filtered spectral propagator

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$28

and the low-pass cutoff is chosen by testing norm conservation. With this choice, the kinetic MPO has max bond dimension below $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$29. In two-dimensional runs on $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$30 grids, the nonlinear update $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$31 reaches ranks in the $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$32–$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$33 range for modulated traps at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$34, while the spatial MPO application is truncated back to $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$35 after each full Trotter step. The principal diagnostics are per-core ranks, chosen $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$36, norm $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$37, energy $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$38, and forward–backward infidelity

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$39

The paper also recommends monitoring norm drift $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$40 and optional residuals for adaptive control of $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$41, $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$42, $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$43, and the low-pass cutoff (Niedermeier et al., 6 Jul 2025).

For long-time QTT time integration of advection-dominated PDEs, the central diagnostic problem is rank explosion. The study tracks per-core TT ranks $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$44, discarded singular values, physical monitors such as oscillation frequency and amplitude in the whistler-wave problem, and symmetry checks such as $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$45 in Maxwell simulations. Reported values are strongly method-dependent: SAT RK4 can produce max ranks of about $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$46 at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$47 and $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$48 at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$49, whereas Lax–Wendroff keeps ranks around $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$50 at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$51 and $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$52 at $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$53; qDLR-PS+RK4 stays near $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$54 at high resolution. Artificial dissipation reduces rank growth but introduces amplitude and phase damping. The recommended practice is to use LW for robust explicit runs, qDLR-PS with basis augmentation for larger time steps, and mapping choices such as sequential rather than interleaved orderings when they reduce inter-axis entanglement (Ye, 13 May 2026).

In tailored quantics algorithms for Poisson and Schrödinger problems, diagnostics combine residual norms, energy trends, virial ratios, singular values, and hierarchy consistency. For Poisson, the stopping criterion is $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$55, and coarse-grid averaging restriction is recommended because it preserves total charge. For eigenproblems, convergence is assessed by energy decay, orthogonality to previously found states, and the virial ratio $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$56 for the ground state. The reported problem scales are extreme: up to $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$57 in two dimensions with only $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$58 parameters for the Poisson benchmark, and up to $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$59 total grid points in four-dimensional vibronic calculations. Energy errors of $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$60–$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$61 in $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$62D $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$63 are obtained with $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$64 (Li et al., 10 Apr 2026).

7. Theoretical structure, feasibility, and persistent limitations

Not every prescribed singular-value profile is admissible. In the TT feasibility problem, the relevant objects are the singular spectra of the unfoldings $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$65, and the squared feasible spectra form closed, convex, polyhedral cones. Feasibility decouples across adjacent TT splits, and a pair $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$66 at an interior mode is constrained by degree bounds, Horn/Klyachko-type inequalities, a Ky Fan analogue,

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$67

and a Weyl analogue,

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$68

The paper also gives a linear-programming feasibility test via hives and honeycombs. For QTT, these conditions apply without change to the folded representation (Krämer, 2017).

The error theory for TT-cross likewise imposes nontrivial conditions: well-conditioned intersections, selected rows and columns aligned with dominant singular spaces, and balanced mode splits. Under these conditions, the global Frobenius error grows polylogarithmically in tensor order rather than exponentially; without them, large values of $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$69, $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$70, $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$71, or $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$72 expose unstable crosses and error amplification (Qin et al., 2022).

Across applications, several limitations recur. Tensorization and folding choices are decisive; pre-scaling can severely degrade structure-preserving compression in anomaly detection; group-comparative TT detectors were found slow with unsatisfactory results; local detectors can be prohibitively slow; too small $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$73 or $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$74 can admit noise artifacts in QTTD; and long-time QTT integration can become noise-dominated even when the underlying PDE discretization is stable (Ali et al., 2024, Rohshap et al., 15 Jul 2025, Ye, 13 May 2026).

Normalization is another recurring issue. Standard TT does not intrinsically enforce $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$75, whereas normalized TT defines the fixed-rank manifold

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$76

and uses the retraction

$X \in \mathbb{R}^{n_1\times \cdots \times n_d}$77

This replaces naive renormalization by a rank-preserving normalization procedure and makes $X \in \mathbb{R}^{n_1\times \cdots \times n_d}$78 a built-in diagnostic target (Peng et al., 6 Nov 2025).

Taken together, these results define QTT diagnostics as a layered discipline rather than a single metric. At the lowest level are bond dimensions, singular values, and orthogonality; at the algorithmic level are truncation tolerances, cross-conditioning, and residuals; at the application level are anomaly scores, entanglement peaks, conservation laws, virial ratios, and forward–backward fidelities. The common principle is that QTT does not merely compress high-resolution objects: it turns their multiscale structure into explicitly measurable quantities.