The future of quantum maritime logistics is now

Voyage Matrices · TTDECOM

Download paper

Kryptur Research presents the first end-to-end Tensor-Train compression and quantum encoding of multi-port voyage-cost tensors — 7,528× GPU compression, 50× classical MIP speedup, and 0.9936 fidelity on a 156-qubit superconducting processor.

10.5281/zenodo.20750317Raja RamGitHub
TT compression
7,528×
1.87 GB → 243 KB on H200 GPU
GPU TT-SVD time
32.3 s
4-D voyage tensor · rank 4 cores
QPU encoding fidelity
0.9936
3-qubit amplitude encoding on 156-qubit device
Classical MIP speedup
50×
<0.2% optimality gap on compressed data

Research highlights

TT Voyage Compression: Tensor-Train Decomposition and Quantum Encoding of Multi-Port Voyage Matrices for Maritime Logistics

Research Paper
Raja Ram M · Kryptur Quantum R&D · Muskan S · Vipul Jain · Kalinga Swain

We compress a 1.87 GB 4-D voyage-cost tensor spanning 15,000 ports, 30 cargo types, 104 weeks, and 5 vessel categories by 7,528× using GPU-accelerated TT-SVD, amplitude-encode the TT cores on a 156-qubit QPU with 0.9936 fidelity, and execute eight quantum algorithms — kernel estimation, QAOA, PCA, swap-test anomaly detection, ensemble VQE, quantum SVM, reservoir forecasting, and ZNE mitigation.

MaritimeTensor-TrainIBM Quantum

Compression ratio: original 1.87 GB vs. 243 KB TT cores

Figure 1
7,528× reduction in 32.3 seconds on H200
GPUTT-SVD

Quantum encoding fidelity: simulator vs. real QPU

Figure 4
F_sim = 1.0 · F_QPU = 0.993647
Amplitude EncodingNISQ
Explore TTDECOM pipeline on GitHubRead more research on our hub

Results & figures

Table 1. GPU compression metrics
MetricValue
Original size (dense)1.87 GB
Compressed size (TT cores)242.8 KB
Compression ratio7,528×
Decomposition time32.3 s
Relative Frobenius error (rank 4)0.359
Pareto optimal rank5 (error 0.064)
Table 2. Quantum algorithm suite on compressed TT cores
AlgorithmQubitsKey result
Amplitude encoding3F = 0.993647 on 156-qubit QPU
Quantum kernel estimation2–410/10 kernel entries measured
QAOA route sampling4Cost Hamiltonian from TT-derived vector
Quantum PCA (VQE)2Largest eigenvalue from G1 covariance
Swap-test anomaly detection3Low-overlap ports flagged
Ensemble VQE220-trial robust energy estimate
Quantum SVM2Kernel matrix → classical SVM
Quantum reservoir forecasting3Linear readout on G2 time series
Compression ratio visualisation: original 1.87 GB vs. 243 KB TT cores
Figure 1. Compression ratio visualisation: original dense voyage tensor (1.87 GB) vs. compressed TT cores (242.8 KB) — a 7,528× reduction in 32.3 seconds on an H200 GPU.
Rank-error Pareto frontier with elbow at rank 5
Figure 2. Rank-error Pareto frontier for TT-SVD decomposition. Elbow at rank 5 yields 6.4% relative error while maintaining exponential compression.
Original vs. reconstructed tensor slice for ports by weeks
Figure 3. Original vs. reconstructed tensor slice (ports × weeks) from the CPU prototype, confirming that TT cores preserve the underlying trade-lane structure.
Fidelity comparison: simulator 1.0 vs. real QPU 0.9936
Figure 4. 3-qubit amplitude encoding fidelity: ideal simulator (F = 1.0) vs. measured real-hardware fidelity (F = 0.993647) on a 156-qubit superconducting processor.
Quantum kernel heatmap compared with classical RBF kernel
Figure 5. Quantum kernel matrix (ZZFeatureMap) compared with classical RBF kernel on port feature vectors extracted from the first TT core.
QAOA route sampling distribution from TT-derived cost Hamiltonian
Figure 6. QAOA route sampling: 1-layer QAOA circuit with cost Hamiltonian constructed from the TT-derived cost vector for maritime route optimisation.
Quantum reservoir computing forecast from G2 core time series
Figure 7. Quantum reservoir computing: fixed random 3-qubit circuit driven by G2 core time series with linear readout trained for voyage-cost forecasting.
Quantum SVM ROC curve for anomaly classification
Figure 8. Quantum SVM: ROC curve for anomaly classification using the quantum kernel matrix fed into a classical support vector machine.
Swap test anomaly detection comparing port states
Figure 9. Swap-test anomaly detection: overlap comparison of port states against a reference, flagging low-overlap anomalies in the compressed voyage tensor.
VQE simulation convergence for quantum PCA
Figure 10. VQE simulation convergence for quantum PCA: largest eigenvalue of the G1 core covariance matrix mapped to a 2-qubit Hamiltonian.
GPU scenario simulation for port closure disruption
Figure 11. GPU scenario simulation: fleet cost impact (+3.69%) under port-closure disruption, computed in 63 seconds on compressed TT cores.

Full paper, pipeline & data

Research Paper: https://doi.org/10.5281/zenodo.20750317 · Open access · TTDECOM Research Initiative · Kryptur OU

Research Paper

Stay up to date

with the latest from

Kryptur Research on

LinkedIn

Tools + Code

Research Paper

TT Voyage Compression technical report — GPU TT-SVD, amplitude encoding, eight quantum algorithms, and 46-stage maritime logistics campaign.

View project →

TTDECOM Pipeline

Tensor-Train decomposition scripts, QPU experiments, FastAPI REST endpoint, and interactive research dashboard.

View project →

Ocean Temperature

Amplitude-encoded global SST on ibm_fez — Tucker compression and real-hardware climate validation.

View project →

Olfactory Vibration

Quantum-biology IETS spectroscopy and vibrational olfaction research.

View project →