1 Basic definitions and notation

1.1 Determinant of a square matrix

Let \(A\in \mathbb{F}^{n\times n}\) with \(\mathbb{F}=\mathbb{R}\) or \(\mathbb{C}\). The determinant \(\det(A)\) is the scalar multilinear alternating function of the columns of \(A\). It satisfies \(\det(AB)=\det(A)\det(B)\), \(\det(A^T)=\det(A)\), and \(\det(A)\neq 0\) exactly when \(A\) is invertible. Determinant inequalities are statements that bound, compare, or estimate determinants under assumptions on \(A\) or related matrices.

1.2 Sign, magnitude, and determinant conventions

Over \(\mathbb{R}\), determinants may be negative; many determinant inequalities control \(\det(A)\) or \(\det(A)\) for positive (semi)definite matrices where \(\det(A)\ge 0\). Over \(\mathbb{C}\), the determinant is complex in general only for non-Hermitian matrices; inequalities are typically phrased for Hermitian or for quantities such as \(\det(A)\) that are real. Conventions vary on whether one works with determinants directly or with logarithms (e.g., \(\log\det\) for positive definite matrices).

Several basic identities motivate common inequality patterns. For invertible \(A\), \(\det(A^{-1})=1/\det(A)\). For products, multiplicativity gives a natural template: \[ \det(AB)=\det(A)\det(B), \]

which can be combined with bounds on each factor or with inequalities for related transforms. For transpose and conjugate transpose, \(\det(A^*)=\overline{\det(A)}\); hence \(\det(A)^2=\det(A^*A)\), a relation often used to connect determinants to singular values.

1.4 Eigenvalues and singular values as determinant inputs

If \(A\) is diagonalizable with eigenvalues \(\lambda_1,\dots,\lambda_n\), then \(\det(A)=\prod_{i=1}^n \lambda_i\). For general \(A\), singular values \(\sigma_1\ge \cdots \ge \sigma_n\ge 0\) satisfy \[

\det(A)=\prod_{i=1}^n \sigma_i,

\]

since \(\det(A^*A)=\det(A)^2=\prod_i \sigma_i^2\). Many determinant inequalities become comparisons between \(\prod \lambda_i\) or \(\prod \sigma_i\) and other spectral or norm-based quantities.

2 Core inequality themes

2.1 Volume interpretation of determinants

Determinants encode “oriented volume scaling.” When \(A\) maps the unit cube (or parallelepiped) to another parallelepiped in \(\mathbb{R}^n\), \(\det(A)\) is the volume factor. Consequently, determinant inequalities often translate to geometric bounds, such as comparing the volume of an image set to the product of certain lengths (as in Hadamard-type statements).

2.2 Lower and upper bounds

Inequalities may provide upper bounds (e.g., controlling \(\det(A)\) by quantities derived from the entries or norms) or lower bounds (e.g., estimating \(\det(A)\) from eigenvalue constraints). The direction depends on the structural assumptions: bounding from above is common for arbitrary matrices via entrywise norms; bounding from below is typical for positive definite matrices with spectral or diagonal constraints.

2.3 Multiplicative inequalities for products

Because determinants multiply, many results are multiplicative in nature: one may compare \(\det(AB)\) to expressions involving singular values, or bound \(\det(A+B)\) by using eigenvalue interlacing. Even when a specific inequality is not directly multiplicative, the underlying determinant structure makes product-related techniques frequent.

2.4 Inequalities under scaling and normalization

Scaling \(A\mapsto tA\) multiplies the determinant by \(t^n\): \(\det(tA)=t^n\det(A)\). Many inequalities normalize matrices (e.g., to fix trace or to standardize diagonal entries) so that comparisons are meaningful across different scales. Log-determinant inequalities are also scale-sensitive in a controlled way: \[ \log\det(tA)=n\log t+\log\det(A). \]

3 Classic determinant inequalities

3.1 Hadamard’s inequality

3.1.1 Statement for general matrices

Hadamard’s inequality bounds the absolute value of the determinant by the product of Euclidean norms of columns (equivalently rows). If \(A=[a_1\,\cdots\,a_n]\) with column vectors \(a_i\in\mathbb{F}^n\), then \[

\det(A)\le \prod_{i=1}^n \|a_i\|_2.

\]

A common corollary for real or complex matrices expresses the bound in terms of diagonal entries of \(A^*A\), since \(\|a_i\|_2^2\) equals the diagonal element \((A^*A)_{ii}\).

3.1.2 Equality conditions and geometric meaning

Equality holds precisely when the columns are pairwise orthogonal in the real case (or orthogonal in the Hermitian inner product sense). Geometrically, orthogonality makes the volume of the parallelepiped equal to the product of edge lengths, with no “shearing loss.” This equality characterization is frequently used to identify extremizers in related optimization problems.

3.2 Determinant bounds via eigenvalues

3.2.1 Trace–determinant relationships for special classes

For positive semidefinite matrices, determinants can be controlled using eigenvalues \(\lambda_i\ge 0\). If \(\operatorname{tr}(A)=\sum_i \lambda_i\) is fixed and \(A\succeq 0\), then the determinant satisfies \[ \det(A)=\prod_{i=1}^n \lambda_i \le \left(\frac{\sum_i \lambda_i}{n}\right)^n, \] by the arithmetic–geometric mean inequality. Thus, for such classes one obtains an upper bound on \(\det(A)\) from trace.

3.2.2 Inequalities for positive semidefinite matrices

More refined inequalities relate \(\det(A)\) to principal minors and to spectral means. For instance, for \(A\succeq 0\), \(\det(A)\) equals the product of eigenvalues, so constraints like \(\lambda_{\max}(A)\le M\) yield \[ \det(A)\le M^{n-1}\lambda_{\min}(A), \] and similar forms. In practice, many determinant bounds under positive semidefiniteness are obtained by combining the product structure with convexity or majorization properties of the eigenvalues.

3.3 Sylvester’s determinant theorem and consequences

Sylvester’s determinant theorem states that for matrices of compatible sizes \(U\in \mathbb{F}^{n\times m}\) and \(V\in \mathbb{F}^{m\times n}\), \[ \det(I_n+UV)=\det(I_m+VU). \] This identity allows determinant inequalities to be transferred between different dimensions, and it underpins many estimates for determinants of rank-updated or Schur-complement-related matrices. Consequences include determinant formulas used in statistics and numerical linear algebra, where \(I+UV\) structures appear naturally.

4 Determinant inequalities under structural assumptions

4.1 Positive definite and positive semidefinite matrices

4.1.1 Monotonicity with respect to the Loewner order

For Hermitian matrices, the Loewner order \(A\succeq B\) means \(A-B\) is positive semidefinite. For positive definite matrices, the log-determinant is concave, and monotonicity of \(\det(\cdot)\) can be expressed through eigenvalue comparisons. While \(\det(A)\) is not globally monotone in every sense, inequalities derived from the ordering and from concavity of \(\log\det\) yield consistent bounds in many standard settings.

4.1.2 Determinant inequalities for principal submatrices

If \(A\succeq 0\), then any principal submatrix \(A_{S,S}\) is also positive semidefinite. Determinants of principal submatrices can be bounded relative to \(\det(A)\) using eigenvalue interlacing and properties of principal minors. Such results are useful when reducing problems to smaller blocks or when analyzing partial information.

4.2 Symmetric and Hermitian matrices

For real symmetric or complex Hermitian matrices, eigenvalues are real and the determinant becomes the product of real eigenvalues. Many inequalities simplify because spectral decomposition exists: one can express \(\det(A)\) as \(\prod_i \lambda_i\) and apply inequalities on \(\lambda_i\). Hermitian structure also aligns determinants with quadratic forms, often enabling proofs that use interlacing, majorization, or convexity of spectral functions.

4.3 Rank constraints and reduced determinants

When \(A\) has rank \(r<n\), its determinant is zero. To capture nontrivial information, one considers reduced determinants such as products of nonzero singular values or pseudo-determinants. Determinant inequalities in this context compare products of the positive singular values to other quantities, including norms and spectral gaps, effectively extending the determinant viewpoint to lower-rank regimes.

4.4 Block matrix determinant bounds

Determinants of block matrices can be expressed using the Schur complement when appropriate invertibility holds. For \[ \begin{pmatrix} A & B\\ C & D \end{pmatrix}, \] if \(A\) is invertible, then \[ \det\begin{pmatrix} A & B\\ C & D \end{pmatrix} =\det(A)\,\det(D-CA^{-1}B). \] Bounds for \(\det\) of the whole matrix can then be obtained by estimating the determinant of the Schur complement, often using positive semidefinite comparisons or singular value inequalities.

5 Inequalities involving norms and condition measures

5.1 Bounds using singular values

5.1.1 Determinant versus operator norm

Singular values relate determinant magnitude to norms. Since \(\det(A)=\prod_{i=1}^n \sigma_i\) and \(\sigma_1=\|A\|_2\) (operator norm),

\[

\det(A)\le \|A\|_2^n,

\]

with equality only in special cases where all singular values match \(\|A\|_2\). This provides a simple upper estimate valid for all matrices.

5.1.2 Determinant versus Frobenius norm

The Frobenius norm obeys \(\|A\|_F^2=\sum_{i=1}^n \sigma_i^2\). Using inequalities between the geometric mean of \(\{\sigma_i^2\}\) and their arithmetic mean yields

\[

\det(A)^{2/n}=\left(\prod_{i=1}^n \sigma_i^2\right)^{1/n}\le \frac{1}{n}\sum_{i=1}^n \sigma_i^2=\frac{\|A\|_F^2}{n},

\] so \[

\det(A)\le \left(\frac{\|A\|_F^2}{n}\right)^{n/2}.

\] These bounds are frequently used when only global norm information is available.

5.2 Condition number implications (when determinant is small)

For invertible \(A\), the smallest singular value satisfies \(\sigma_n=\det(A)/\prod_{i=1}^{n-1}\sigma_i\), so a small determinant indicates that at least one singular value is small. In numerical analysis, the condition number \(\kappa_2(A)=\sigma_1/\sigma_n\) grows as \(\sigma_n\) decreases, linking determinant-based measures to sensitivity of solutions of linear systems.

5.3 Determinant estimates from perturbations

Determinants change continuously with matrix entries, and bounds can be expressed via perturbation theory. If \(A\) is invertible and \(E\) is a perturbation, one often studies \(\det(A+E)\) through \[ \det(A+E)=\det(A)\,\det(I+A^{-1}E), \] and then bounds \(\det(I+X)\) in terms of norms or eigenvalues of \(X\). Such estimates are central to stability analysis where determinants act as proxies for volume and invertibility robustness.

6 Determinant inequalities and matrix transformations

6.1 Similarity transformations and invariance

For invertible \(P\), determinants are invariant under similarity: \[ \det(P^{-1}AP)=\det(A). \] Thus, determinant inequalities that depend only on the determinant itself are similarity-invariant. However, when inequalities involve additional spectral constraints (eigenvalues, norms, or definiteness), similarity invariance may not directly preserve those constraints unless the transformation is chosen appropriately (e.g., unitary similarity preserves eigenvalues and Hermitian structure).

6.2 Effect of congruence transformations

For congruence with invertible \(P\), \(\det(P^*AP)=\det(P^*)\det(A)\det(P)=\det(P)^2\det(A)\) in the complex case. Congruence changes determinant magnitude in a controlled multiplicative way while preserving definiteness: if \(A\succeq 0\), then \(P^*AP\succeq 0\). This makes congruence transformations a common tool for normalizing problems and transferring inequalities between matrices with comparable structure.

6.3 Determinant inequalities for Schur complements

Schur complements connect determinant comparisons to inequalities for positive semidefinite matrices. When \(A\succeq 0\) and a block matrix is positive semidefinite, the Schur complement is also positive semidefinite, enabling determinant bounds via eigenvalue or trace arguments applied to the Schur complement. This framework turns block determinant questions into single-matrix determinant inequalities.

6.4 Inequalities for affine combinations

For positive definite matrices, the log-determinant function is concave in the matrix argument. As a result, affine combinations satisfy inequalities of the form \[ \log\det(tA+(1-t)B)\ge t\log\det(A)+(1-t)\log\det(B), \quad 0\le t\le 1, \] leading to multiplicative bounds on \(\det(\cdot)\) after exponentiation. Such statements underpin convex optimization models involving \(-\log\det\) or \(\log\det\) objectives.

7 Equality cases and extremal principles

7.1 Characterizing when bounds are tight

Equality conditions are central in determinant inequalities because they reveal the geometric or spectral structure producing extremal behavior. In Hadamard’s inequality, orthogonality of columns yields equality. In trace–determinant bounds for positive semidefinite matrices, equality typically requires all eigenvalues to be equal (under fixed trace), corresponding to scalar multiples of the identity.

7.2 Extremizers for Hadamard-type inequalities

For inequalities controlled by column lengths or row lengths, extremizers often correspond to configurations where the relevant vectors become mutually orthogonal, maximizing the product of independent contributions. When combined with additional constraints (e.g., fixed diagonal entries), the extremizers can become diagonal or nearly diagonal after appropriate normalization.

7.3 Minimization/maximization frameworks

Many determinant inequalities appear as special cases of broader extremal principles: maximizing \(\log\det(A)\) subject to linear constraints often yields solutions with structured eigenvalues (e.g., “water-filling” phenomena in communications). Conversely, minimizing determinants subject to positivity constraints commonly produces boundary cases where rank drops or eigenvalues become equalized. These frameworks link determinant inequalities to convexity and to optimization geometry.

8 Applications in algebra and linear algebra

8.1 Systems of linear equations and stability

In solving \(Ax=b\), invertibility and sensitivity depend on the spectrum of \(A\). Determinant inequalities help estimate when \(A\) is near singular by bounding \(\det(A)\) or relating it to smallest singular values. This supports stability assessments, including qualitative conclusions about error amplification and numerical conditioning.

8.2 Gram matrices and least-squares contexts

Gram matrices \(G=X^*X\) satisfy \(G\succeq 0\) and \(\det(G)\) equals the squared volume of the parallelepiped spanned by the columns of \(X\). Hadamard’s inequality and eigenvalue-based bounds thus translate into statements about the conditioning of least-squares problems and the geometric arrangement of data vectors.

8.3 Convexity/concavity connections to log-det

The concavity of \(\log\det\) on the cone of positive definite matrices creates a systematic link between determinant inequalities and convex analysis. This connection supports both theoretical bounds (e.g., affine-combination inequalities) and algorithmic design in optimization problems where determinants encode uncertainty volume, entropy-like quantities, or covariance scale.

8.4 Algorithms and computational considerations (overview)

Computing determinants directly can be expensive and numerically unstable for large or ill-conditioned matrices. Practical algorithms often rely on factorizations (LU, QR, Cholesky) and on estimating \(\log\det\) via eigenvalues or Cholesky diagonals. Determinant inequalities are used both to bound outputs and to guide regularization strategies, especially when matrices approach singularity.

9 Common proof techniques (algebraic viewpoint)

9.1 Eigenvalue majorization approaches

When a function of eigenvalues is symmetric and suitably convex or concave, majorization tools can produce determinant inequalities. Since \(\det(A)=\prod_i \lambda_i\) for Hermitian \(A\), many proofs reduce to inequalities for products of eigenvalues by comparing eigenvalue distributions across matrices.

Norm relations frequently produce determinant bounds. Since determinants can be expressed through singular values, inequalities between arithmetic means and quadratic means (often derived from Cauchy–Schwarz) lead directly to upper bounds on \(\det(A)\) in terms of \(\|A\|_F\) or other norm measures.

9.3 Jensen-type arguments for log-determinant

For \(A\succ 0\), \(\log\det\) is concave. Proofs of affine-combination inequalities can be framed via Jensen’s inequality applied to the concave matrix functional, yielding determinant comparisons for \(tA+(1-t)B\) without requiring explicit eigenvalue computations.

9.4 Schur complement and matrix determinant lemmas

Block determinant identities reduce complex determinant questions to determinants of smaller matrices. Proofs then combine Schur complements with positivity, interlacing, or norm bounds. Matrix determinant lemmas and rank-update formulas (closely related to Sylvester’s theorem) allow controlled handling of structured perturbations.

10 Worked examples

10.1 Computing and comparing determinants with inequalities

Suppose \(A\in\mathbb{R}^{n\times n}\) has Frobenius norm \(\|A\|_F\). Without computing \(\det(A)\) exactly, one can bound

\[

\det(A)\le \left(\frac{\|A\|_F^2}{n}\right)^{n/2}.

\] This example illustrates how a single global norm quantity yields a concrete comparison for the determinant’s possible magnitude.

10.2 Applying Hadamard to structured matrices

Let \(A\) be formed by columns \(a_i\). Hadamard gives \[

\det(A)\le \prod_{i=1}^n \|a_i\|_2.

\] If \(A\)’s columns are orthogonal (for instance, when \(A\) is diagonal in an orthonormal basis), equality holds. The example demonstrates how geometric structure (orthogonality) becomes algebraic tightness.

10.3 Example bounds from singular values

If \(A\) has singular values \(\sigma_1,\dots,\sigma_n\) and operator norm \(\|A\|_2=\sigma_1\), then

\[

\det(A)=\prod_{i=1}^n \sigma_i\le \sigma_1^n=\|A\|_2^n.

\]

If additionally \(\sum_{i=1}^n \sigma_i^2=\|A\|_F^2\) is known, one can use the Frobenius-based estimate to tighten the bound, showing how different norm inputs lead to distinct determinant upper bounds.

10.4 Example with a block matrix and a Schur complement

Consider a block matrix \[ M=\begin{pmatrix} A & B\\ C & D \end{pmatrix} \] with \(A\) invertible. The determinant factorizes as \[ \det(M)=\det(A)\,\det(D-CA^{-1}B). \] If one knows (for instance) that the Schur complement \(S=D-CA^{-1}B\) is positive semidefinite and has trace bounded, then trace–determinant inequalities on \(S\) give \[ \det(S)\le \left(\frac{\operatorname{tr}(S)}{k}\right)^k \] for \(k\) equal to the size of \(S\) (or an adjusted dimension if \(S\) is reduced). Multiplying by \(\det(A)\) yields an explicit upper bound for \(\det(M)\) without direct determinant computation of the full block matrix.