tensor methods, Oil & Gas Glossary

Tensor methods are the family of mathematical techniques that represent directionally dependent reservoir properties as rank-two tensors rather than as single scalar values, allowing geologists and reservoir engineers to capture the fact that quantities like permeability, stress, and dispersion behave differently along each principal axis of a rock. In three dimensions a permeability tensor has nine components arranged in a 3x3 matrix, with three independent diagonal terms (kxx, kyy, kzz) describing flow along the coordinate axes and six off-diagonal terms describing cross-flow that occurs when fluid pressure gradients and the resulting flow vectors are not aligned. The tensor is symmetric for physically realisable rocks, so only six of those nine numbers are independent, but those six numbers carry far more information than the single Darcy permeability commonly reported on a core plug. When the coordinate system is rotated the components change in a defined way, and the eigenvalues of the matrix give the three principal permeabilities and their orientations, which is the way the rock actually wants to conduct fluid. This matters enormously when an interpreter is upscaling a million-cell fine-scale geological model into a 50,000-cell simulation grid for use in a full-field reservoir simulator, because the simple harmonic and arithmetic averaging used for scalar permeability throws away the directional information that drives early water breakthrough, channelised flow along fractures, and aliasing between layered facies. In the Western Canadian Sedimentary Basin the directional contrast can exceed two orders of magnitude in the Montney siltstone (where bedding-parallel kh dominates kv by 50 to 200 times), in the Duvernay shale (where natural fracture networks orient flow along the regional maximum horizontal stress direction at 035 to 045 degrees azimuth), and in the McMurray oil sands of the Athabasca region (where inclined heterolithic stratification creates baffle networks that respond differently to SAGD steam rising vertically than to producer drainage flowing horizontally). Tensor methods are therefore central to reservoir simulation studies under AER Directive 083 thermal in-situ approval submissions, to history-matched performance forecasts for ARC Resources Montney developments, and to fracture-network upscaling in Duvernay simulation work conducted by operators including Chevron Canada, ConocoPhillips Canada, and CNRL.

Key Takeaways

Nine-component permeability matrix: A full 3D permeability tensor stores nine components in a 3x3 matrix, of which six are independent because the matrix is symmetric. The three diagonal terms describe flow along coordinate axes, the off-diagonals describe cross-flow when pressure gradient and flow vector are misaligned. Scalar Darcy values throw all six off-diagonal terms away, which is why simulator output diverges from observed pressure response in fractured and layered WCSB plays.
Upscaling preserves direction: Tensor upscaling techniques (flow-based, periodic boundary, and Oda methods) compute equivalent coarse-cell permeability by simulating steady-state flow through the underlying fine grid and back-solving the tensor that reproduces that flow. This preserves the principal-axis orientation and magnitude, so a coarse-cell Montney bench keeps its 50 to 200 kh/kv anisotropy ratio instead of being smeared to an arithmetic mean of about 8 to 12 millidarcy.
Fracture network handling: In Duvernay and Horn River fractured plays the Oda tensor sums each natural-fracture set as a contribution to the bulk tensor, weighted by aperture cubed and orientation. The resulting permeability ellipsoid is highly elongated along the regional maximum horizontal stress azimuth, typically 035 to 045 degrees in the Kaybob and Edson areas. Simulators that ignore the tensor will mis-predict frac hit timing between adjacent ARC Resources or CNRL pads by months.
MPFA grid coupling required: Conventional two-point flux approximation (TPFA) solvers cannot correctly handle off-diagonal tensor components on non-K-orthogonal grids. Multi-point flux approximation (MPFA) discretisation is required, which is why CMG GEM, Schlumberger INTERSECT, and tNavigator all expose MPFA options. Using TPFA with a full tensor on a non-orthogonal corner-point grid produces spurious diagonal flows that can misroute injection water by tens of metres per year of simulation time.
Stress and dispersion tensors too: Tensor methods extend beyond permeability. The geomechanical stress tensor (six independent components) drives fracture orientation in Montney and Duvernay completions design, and the hydrodynamic dispersion tensor governs solvent mixing in vapour-extraction and solvent-assisted SAGD designs evaluated by Cenovus and Imperial Oil at Foster Creek, Christina Lake, and Cold Lake.

Flow-Based Upscaling Workflow for Montney Simulation

A typical Montney upscaling workflow at a major WCSB operator starts with a static fine-scale model from Petrel containing approximately 4 to 8 million cells at 50 m by 50 m by 0.5 m resolution. Each cell carries a scalar absolute permeability from the petrophysical model. The geomodeller then groups cells into coarse simulation blocks of 250 m by 250 m by 5 m and runs a steady-state single-phase flow solver across each block in turn, with periodic boundary conditions applied on opposing faces. The solver computes the flow response to a unit pressure gradient applied along x, y, and z in successive runs. The resulting flux vectors are inverted to give the six independent components of the equivalent permeability tensor for that coarse block. The result is written back into the simulation grid as a full tensor field, which CMG IMEX or GEM then ingests via the *KX, *KY, *KZ, *KXY, *KXZ, *KYZ keywords.

Cost and Run-Time Tradeoffs in WCSB Workflows

Tensor upscaling is computationally expensive. A 4-million-cell Montney static model upscaled to a 50,000-cell simulation grid using flow-based tensor methods typically takes 6 to 18 hours on a 32-core workstation, compared to roughly 5 minutes for harmonic averaging. Software licences for the upscaling modules (Schlumberger Petrel RE, CMG CMOST, Emerson Roxar) run CAD 75,000 to CAD 220,000 per year per seat. For a Duvernay full-field simulation project supporting an AER Directive 071 thermal monitoring submission, the total upscaling and simulation effort typically consumes 200 to 400 person-hours of reservoir engineering time and CAD 80,000 to CAD 250,000 in software and compute, but it reduces the standard deviation of 10-year recovery forecasts by 30 to 50 percent compared to scalar upscaling.

Fast Facts

The mathematical framework that lets reservoir engineers turn millions of core-plug permeability measurements into a few dozen tensors traces back to a 1985 SPE paper by Begg, Carter, and Dranfield that introduced flow-based upscaling for North Sea reservoirs. By 2024 the same arithmetic, accelerated on Nvidia A100 GPUs, can upscale a 60-million-cell McMurray oil sands geological model into a 100,000-cell SAGD simulation grid in under 45 minutes, a calculation that would have required several weeks on a Sun workstation in the early 1990s.

Tensor methods rarely stand alone in a reservoir engineer's toolbox. The Permeability entry covers the underlying scalar concept and the Darcy unit conventions that tensors generalise, while Anisotropy describes the directional rock behaviour that makes the tensor representation necessary in the first place. The Reservoir Simulation entry explains the finite-volume numerical framework in which tensors are consumed, and Upscaling covers the broader family of techniques for moving from fine geological grids to coarse simulation grids while preserving flow behaviour.

Duvernay Frac Hit Modelling: Tensor Methods in Practice

A 2024 simulation study by an Edson-area Duvernay operator (a sub-licensee of Chevron Corporation) used full-tensor upscaling to forecast inter-well frac hit timing between an existing 2.4 km lateral and three new infill wells planned at 250 m spacing. The fine-scale discrete fracture network model contained 18,000 natural fractures binned into three orientation sets. Oda tensor upscaling produced principal permeabilities of 0.42 mD along the maximum horizontal stress azimuth (042 degrees) and 0.018 mD perpendicular, an anisotropy ratio of 23 to 1. Coarse-grid simulation in CMG GEM predicted first frac hits at 14 to 21 days post-stimulation. Compute cost for the upscaling step alone was CAD 38,000.

Field results confirmed first measurable pressure response at the parent well within 16 days of toe-stage stimulation on the middle infill, validating the tensor approach. The operator subsequently re-spaced two additional planned pads from 250 m to 350 m, deferring approximately CAD 14.2 million in capital while preserving the same total recoverable resource based on the tensor-informed drainage geometry.