Does a shear preserve parallel lines?

Yes, a nonsingular linear shear maps parallel lines to parallel lines.

Does it preserve right angles?

Generally no; a rectangle commonly becomes a nonrectangular parallelogram.

Can every shear be written with the same horizontal matrix?

Only after choosing suitable coordinates. A shear fixing a differently oriented line has a different matrix in the original coordinate system.

AI Glossary / Mathematics and optimization / Beginner

Shear matrix

Definition

A shear matrix represents a transformation that slides points parallel to a fixed line or plane by amounts proportional to their distance from it. A simple two-dimensional horizontal shear maps (x, y) to (x + ky, y). With column-vector coordinates, its matrix is [[1, k], [0, 1]]. The parameter k controls the amount and direction of the slant.

What it means

A shear can turn a rectangle into a parallelogram without changing its area. Points on the fixed axis remain in place, while points farther from that axis move farther. This differs from translating every point by the same displacement, rotating through a fixed angle or scaling all distances in one direction. Shears are useful building blocks in geometry, graphics and numerical linear algebra. They can also appear in image transformations used to prepare or augment machine-learning data.

How it works

Multiply each point's coordinate vector by the matrix. For a horizontal shear, the new x coordinate adds k times y, while y stays unchanged. A vertical shear instead uses [[1, 0], [k, 1]], giving (x, y + kx). The determinant of either simple shear matrix is one, so oriented area is preserved and an inverse exists. Reverse a simple shear by negating k. A shear about a line away from the origin requires a translation to that line's coordinate frame and back, producing an affine transformation rather than only a two-by-two linear map. Matrix and coordinate conventions must be kept consistent.

Three practical examples

A graphics application can slant a vector shape while retaining straight edges and parallel lines.
A document-image experiment can apply controlled shear to simulate some forms of slant, then test whether a recognition model remains accurate on appropriate examples.
An introductory mechanics model can represent one layer sliding relative to another with a simple shear, while leaving actual material forces and constitutive behavior to a separate physical model.

Think of it this way

Think of sliding horizontal strips of a drawing so that higher strips move farther sideways. The bottom strip stays fixed, and the picture develops a slant. The analogy matches a horizontal shear when strip displacement changes linearly with height. Arbitrary strip movements, bending and uneven displacement do not necessarily form one shear matrix, and physical material behavior is not determined by the geometry alone.

Worked example

Take the rectangle with vertices A = (0, 0), B = (2, 0), C = (2, 1) and D = (0, 1). Apply the horizontal shear with k = 1.5. The transformed vertices are A′ = (0, 0), B′ = (2, 0), C′ = (3.5, 1) and D′ = (1.5, 1). The base remains length 2 and the perpendicular height remains 1, so the parallelogram's area is still 2. Its formerly vertical side now has length √(1.5² + 1²) = √3.25, approximately 1.803, showing that area preservation does not imply length preservation. Its axis-aligned bounding box has width 3.5 and height 1, so that box's area is 3.5 even though the shape's area is 2. To reverse the transformation of C′, calculate x = 3.5 − 1.5(1) = 2 and y = 1. This recovers C exactly in real arithmetic. If a raster image is clipped to the original two-unit-wide canvas, some transformed content would lie outside the canvas despite the geometric inverse existing.

Illustrative case study

In a fictional project, Maple Forms tests a document-recognition pipeline with synthetically slanted training images. The team transforms annotations with the same geometry, expands the canvas when needed and checks whether labels remain readable. It limits shear strength to changes relevant to the intended documents. Evaluation uses separate original scans so that success on synthetic transformations is not mistaken for proven performance on every scanning distortion. The case illustrates data preparation without inventing an accuracy improvement.

Common misconceptions and mistakes

A shear is not generally a rotation: lengths and angles usually change. Determinant one is necessary for the simple area-preserving shear described here, but it is not sufficient to identify a shear; many other transformations also have determinant one. Matrix multiplication order matters when combining horizontal and vertical shears or other transformations. Setting two off-diagonal entries independently in [[1, a], [b, 1]] gives determinant 1 − ab, so that matrix is not generally a simple area-preserving shear. Raster interpolation, clipping and rounding can lose information even when the underlying continuous map is invertible.

Business use

Use shear when the intended geometric change is a position-dependent slant along a fixed direction. Specify whether coordinates are row or column vectors, where the origin lies and whether image y coordinates increase upward or downward. Choose interpolation and boundary handling deliberately for images, and transform masks, landmarks and annotations consistently. Validate that the operation preserves the meaning of training labels. For exported designs, inspect the transformed canvas and bounding boxes instead of assuming area preservation prevents cropping or layout problems.

Check your understanding

A unit-area square is sheared and its bounding box grows. Has the shear necessarily changed the square's area?

Answer: No. The transformed parallelogram still has unit area under a simple shear, while its axis-aligned bounding box may contain extra empty space.

Related terms

Linear transformation
Affine transformation
Determinant
transformation matrix
image augmentation
homogeneous coordinates

Learn next

draw the worked rectangle before and after transformation, then apply the inverse matrix. Compare a horizontal shear followed by a vertical shear with the reverse order on the point (1, 1). Use the results to see why transformation order belongs in a reproducible graphics pipeline.

Inverse matrix

Sources

Sources: TU Delft, Linear Algebra, section 3.3.4, Shear Transformations, https://interactivetextbooks.tudelft.nl/linear-algebra/Chapter3/GeometryofLinearTransformations.html . The rectangle, inverse calculation and fictional document-processing workflow are original examples using the stated column-vector convention.

Previous: Shaving Next: Shi - Tomasi (good features to track)