Can a determinant be negative?

Yes; for real transformations, that indicates reversed orientation.

Does determinant one mean the matrix is the identity?

No. The diagonal example has determinant one but very different directional scaling.

Does zero determinant imply noninvertibility?

Yes in exact arithmetic for a square matrix; computed results require numerical interpretation.

AI Glossary / Mathematics and optimization / Advanced

Determinant

Definition

The determinant is a scalar associated with a square matrix. For a real matrix representing a linear transformation, its absolute value gives the factor by which full-dimensional volumes change, while its sign indicates whether orientation is preserved or reversed. A zero determinant means that the transformation collapses at least one dimension and the matrix is not invertible. Determinants connect algebraic properties of matrices with geometric changes in space.

What it means

In two dimensions, the relevant volume is area; in three dimensions, it is ordinary volume. A determinant of five means that areas or volumes are multiplied by five, depending on the dimension. A determinant of minus five gives the same magnitude of scaling but reverses orientation.
This is a combined measure of the transformation's effect across dimensions. It does not reveal every individual stretch, nor does it measure a model's accuracy. Two matrices can have the same determinant while behaving very differently when used to solve a numerical problem.

How it works

For a two-by-two matrix with rows (a, b) and (c, d), the determinant is ad − bc. The identity matrix has determinant one. Swapping two rows or two columns reverses the sign; multiplying one row by a scalar multiplies the determinant by that scalar. Adding a multiple of one row to another leaves it unchanged.
For larger matrices, numerical software typically uses a factorization rather than expanding all terms of the determinant formula. Determinants multiply under composition: det(AB) = det(A)det(B). In exact arithmetic, a square matrix is invertible precisely when its determinant is nonzero. Finite-precision calculations require additional care when assessing near-singularity.

Three practical examples

An image transformation stretches, shears or reflects coordinates. Its determinant describes the area scaling and orientation change for the linear part of the transformation, helping distinguish a reflection from an ordinary stretch.
A probabilistic model uses a covariance matrix in a multivariate Gaussian density. Its determinant contributes to the normalization associated with the distribution's spread. The covariance structure and other terms remain essential; the determinant alone does not define the density.
A change-of-variables calculation uses the absolute determinant of a Jacobian to adjust local volume. For a nonlinear mapping, that factor can depend on location, so one global area multiplier need not describe the entire transformation.

Think of it this way

Think of drawing a unit square on a flexible sheet and applying a linear transformation. The resulting parallelogram may be stretched or sheared. Its area tells you the magnitude of the determinant. The ordering of its transformed directions tells you the sign. A very long, thin parallelogram can have the same area as a square despite its different shape.

Worked example

Let A have rows (2, 1) and (1, 3). Its determinant is 2 × 3 − 1 × 1 = 5. The first coordinate direction (1, 0) maps to the first column, (2, 1), and the second direction (0, 1) maps to (1, 3).
The unit square therefore maps to a parallelogram with vertices (0, 0), (2, 1), (3, 4) and (1, 3). Its area is five square units, and the positive determinant means orientation is preserved. The matrix is invertible.
Swap A's two columns. The resulting matrix has rows (1, 2) and (3, 1), with determinant 1 × 1 − 2 × 3 = −5. The transformed area is still five, but the ordered directions have reversed orientation.
Now consider B with rows (2, 4) and (1, 2). Its determinant is 2 × 2 − 4 × 1 = 0. Its second column is twice its first, so every transformed point lies on the line spanned by (2, 1). The original square loses its two-dimensional area. Different inputs can now produce the same output, preventing an inverse transformation on the full plane.
These calculations illustrate magnitude, sign and zero separately, without treating a negative determinant as an error or a zero determinant as a negative area.

Illustrative case study

In a fictional analytics project, a team uses the determinant as its only check that a coordinate transformation is numerically well behaved. It compares the identity matrix with a diagonal matrix D whose diagonal entries are 1,000 and 0.001.
Both determinants are one, but D stretches one direction by 1,000 and shrinks the other by 1,000. Its largest-to-smallest singular-value ratio, the spectral condition number, is 1,000/0.001 = 1,000,000. The identity's condition number is one.
The team realizes that equal volume scaling does not imply equal sensitivity when solving associated systems. It adds conditioning checks and examines feature units and scaling. It also avoids using a fixed absolute determinant cutoff as a universal test: rescaling a matrix can change its determinant dramatically without changing the underlying relative conditioning. The determinant remains useful, but it answers a different question from the condition number.

Common misconceptions and mistakes

The ordinary determinant is defined for square matrices, not arbitrary rectangular data tables. A nonzero determinant establishes exact invertibility, not good conditioning. Its size also depends on scale: multiplying every entry of an n-by-n matrix by c multiplies the determinant by cⁿ.
A positive determinant does not establish positive definiteness. For example, the two-by-two diagonal matrix with both diagonal entries −1 has determinant one while its quadratic form is negative for every nonzero real vector. Finally, a determinant reported as zero in floating-point arithmetic can reflect numerical underflow rather than exact mathematical singularity.

Business use

Use determinants when interpreting transformations, probability calculations or volume changes. State the matrix's meaning and units, and distinguish determinant-based questions from conditioning and model-quality questions. When a calculation requires very large or very small determinant magnitudes, a sign and logarithmic absolute determinant can avoid representing the raw magnitude directly. That technique addresses numerical range; it does not make an ill-conditioned problem well conditioned.

Check your understanding

What is det(A) in the example?

Answer: Five.

Why is det(B) zero?

Answer: Its columns are linearly dependent.

Why is D's determinant insufficient as a reliability check?

Answer: Opposing stretches cancel in their product while its condition number remains large.

Related terms

Learn next

Compare determinant, rank and condition number on small matrices. Then study factorization-based computation and the use of logarithmic determinants in probability models.

Condition number

Sources

Aalto Dictionary of Machine Learning, determinant: algebraic properties, volume scaling, orientation and invertibility. https://aaltodictionaryofml.github.io/ADictML_Main.pdf NumPy, numpy.linalg.slogdet: sign and logarithmic magnitude, factorization-based computation and numerical range. https://numpy.org/doc/stable/reference/generated/numpy.linalg.slogdet.html The matrix calculations and fictional conditioning comparison are original teaching illustrations.

Previous: Detection error tradeoff Next: Determiner