The Forms, the Norms and the Singular Set of $M_5(\mathbb{C})$
Introduction
The determinant is a multiplicative form on $M_5(\mathbb{C})$: it is a homogeneous polynomial of degree $5$,
$$ \det(\lambda A)=\lambda^5\det A,\qquad \det(AB)=\det A\det B, $$
and it is the norm of the algebra in the sense of a multiplicative form. It is not a norm in the analytic sense: it is complex-valued, it is not positive, and it is not homogeneous of degree one. The analytic norms are the Frobenius norm
$$ \langle A\rangle_*=\operatorname{tr}(A^*A)=\sum_{i,j}|A_{ij}|^2 $$
and the operator norm
$$ \|A\|=\sup_{\|x\|=1}\|Ax\| . $$
The article distinguishes the determinant form from the two norms, and it studies the singular set
$$ \Sigma=\{A:\det A=0\}, $$
the zero locus of the determinant. The singular set is a complex hypersurface of complex dimension $24$ and real codimension $2$ in $M_5\cong\mathbb{R}^{50}$. It is the union of the determinantal varieties of rank at most $4$, with the rank stratification
$$ M_5^{(0)}\subset M_5^{(1)}\subset M_5^{(2)}\subset M_5^{(3)}\subset M_5^{(4)}=\Sigma, $$
of complex dimensions $0,9,16,21,24$ and complex codimensions $25,16,9,4,1$. The successive differences of the dimensions are $9,7,5,3$, and the codimension of $\Sigma$ in $M_5$ is $1$; the odd numbers
$$ 1,3,5,7,9 $$
are the dimensions of the adjoint summands of the principal $\mathfrak{sl}_2$.
The Hermitian trace form $\operatorname{tr}(A^*B)$, the Frobenius form, is positive definite and has no null cone; the singular set is not a null cone of the metric. The complex bilinear pairing $\operatorname{tr}(AB)$ is indefinite, and its isotropic cone is a different set. The article assumes the matrix algebra, the trace and the involution; it uses the spectral theorem only through the definition of the operator norm, and it proves the basic inequalities directly.
The Determinant as a Form
Homogeneity and Multiplicativity
Proposition. The determinant is a homogeneous polynomial of degree $5$ on $M_5(\mathbb{C})$:
$$ \det(\lambda A)=\lambda^5\det A,\qquad \det(AB)=\det A\det B,\qquad \det I=1 . $$
It is not additive: $\det(A+B)\ne\det A+\det B$ in general.
Proof. Each of the $5!$ terms in the Leibniz formula contains exactly one entry from each of the five rows and columns, so the degree is $5$; the multiplicativity is the standard theorem of determinants; the identity has determinant $1$. For the failure of additivity, take $A=B=I$: $\det(2I)=32\ne2=\det I+\det I$.
Corollary. The determinant is a multiplicative form of degree $5$: it is a polynomial that converts the algebra product into the product of scalars. The zero set is the set of non-invertibles.
Proof. $\det(AB)=\det A\det B$ and $\det A\ne0$ iff $A$ is invertible; the zero set is the complement of the group of invertibles.
The Norm of a Matrix
Definition. The determinant norm of $A$ is $N(A)=\det A$.
Proposition. The determinant norm is multiplicative and homogeneous of degree $5$, but it is not a norm in the analytic sense: it takes complex values, it is not positive, and it is not subadditive.
Proof. The multiplicativity and homogeneity are above. The determinant of $iI$ is $i^5=i$, which is not real; the determinant of a nilpotent matrix is $0$ although the matrix is nonzero; the determinant of $I+(-I)=0$ is $0$, while $\det I+\det(-I)=1-1=0$, but $\det(I+I)=32\ne2$. The triangle inequality fails.
Corollary. The determinant norm is a multiplicative form: $N(AB)=N(A)N(B)$. It is not a quadratic form, so $M_5(\mathbb{C})$ is not a composition algebra in the classical sense (a composition algebra has a nondegenerate quadratic norm and dimension $1,2,4$ or $8$). The form is not positive definite and has nontrivial zeros; the algebra is not a division algebra.
Proof. The multiplicativity is the composition property; the form has degree $5$, not $2$, so it is not the quadratic norm of a composition algebra. The nilpotent matrices are nonzero with norm zero, so there are zero divisors.
Remark. The word "norm" has three meanings in this category. The determinant norm is the multiplicative form of degree $5$; the Frobenius norm is the Euclidean norm of the $25$ coordinates; the operator norm is the sup norm on the unit sphere. They agree only in special cases: for a scalar matrix $\lambda I$, $\det(\lambda I)=\lambda^5$, $\langle \lambda I\rangle_*^{1/2}=\sqrt5|\lambda|$, $\|\lambda I\|=|\lambda|$. The three must not be confused.
The Determinant Form on $M_5$
Definition. The polar forms of the determinant are the multilinear forms obtained by polarizing the homogeneous polynomial $\det$. The first polar form at $A$ in the direction $H$ is the derivative
$$ D\det_A(H)=\operatorname{tr}(\operatorname{adj}(A)H). $$
Proposition. The derivative of the determinant at $A$ is the linear form $H\mapsto\operatorname{tr}(\operatorname{adj}(A)H)$; at $A=I$ it is $H\mapsto\operatorname{tr}H$.
Proof. Expand $\det(A+tH)$ in powers of $t$; the linear term is the sum over the positions of $H$ with the cofactor of $A$, which is $\operatorname{tr}(\operatorname{adj}(A)H)$. At $A=I$, $\operatorname{adj}(I)=I$, so the derivative is the trace.
Corollary. The determinant is a local coordinate near $I$; its level set $\det=1$ is a hypersurface of complex dimension $24$ with tangent space $\mathfrak{sl}_5$.
Proof. The derivative at $I$ is the nonzero linear form $\operatorname{tr}$; its kernel is $\mathfrak{sl}_5$, of dimension $24$; the implicit function theorem gives the smooth hypersurface.
Remark. The determinant is a polynomial, not a norm. The polar forms are the algebraic linearizations; the gradient with respect to the trace pairing is the adjugate, $\langle\operatorname{adj}(A),H\rangle=\operatorname{tr}(\operatorname{adj}(A)H)$. The determinant is the norm form of the algebra, and its polar forms are the algebraic substitutes for the inner product that the determinant does not provide.
The Bilinear and Hermitian Forms
The Trace Pairing
Definition. The trace pairing is $\langle A,B\rangle_{\mathrm{tr}}=\operatorname{tr}(AB)$.
Proposition. The trace pairing is complex bilinear, symmetric and nondegenerate; it is indefinite, with isotropic vectors: a nonzero nilpotent $A$ satisfies $\operatorname{tr}(A^2)=0$.
Proof. The bilinearity and symmetry are the trace identities; the nondegeneracy is the matrix-unit computation $\langle E_{ij},E_{ji}\rangle=1$; a nilpotent $A\ne0$ has $A^2=0$ and trace zero.
Corollary. The trace pairing has an isotropic cone $\{A:\operatorname{tr}(A^2)=0\}$, a complex quadric hypersurface. The cone is not the singular set $\det=0$ and is not the null cone of the Frobenius form.
Proof. The equation $\operatorname{tr}(A^2)=0$ is a nonzero quadratic equation, hence a quadric hypersurface; the nilpotents give nonzero points; the determinant zero set is a different hypersurface.
The Frobenius Form
Definition. The Frobenius form is $\langle A,B\rangle_F=\operatorname{tr}(A^*B)$.
Theorem. The Frobenius form is a positive definite Hermitian inner product; the only null vector is zero. It defines the Frobenius norm $\langle A\rangle_*^{1/2}=\sqrt{\langle A,A\rangle_F}$.
Proof. $\langle A,A\rangle_F=\sum_{ij}|A_{ij}|^2$, positive for $A\ne0$; the conjugate symmetry is $\overline{\operatorname{tr}(A^*B)}=\operatorname{tr}(B^*A)$.
Corollary. The Frobenius form has no null cone. Its unit sphere is the set of matrices with $\sum_{ij}|A_{ij}|^2=1$, a real sphere of dimension $49$.
Proof. Positive definiteness excludes nonzero null vectors; the unit sphere is the sphere in $\mathbb{R}^{50}$.
The Operator Norm
Definition. The operator norm is
$$ \|A\|=\sup_{\|x\|_2=1}\|Ax\|_2, $$
where $\|\cdot\|_2$ is the standard Hermitian norm on $\mathbb{C}^5$.
Proposition. The operator norm is a norm, and
$$ \|A\|\le\langle A\rangle_*^{1/2}\le\sqrt5\,\|A\| . $$
Proof. The operator norm is the sup of the continuous function $x\mapsto\|Ax\|$ on the compact unit sphere; the triangle inequality and homogeneity are immediate. For the first inequality, $\|Ax\|_2^2=\sum_i|\sum_jA_{ij}x_j|^2\le\sum_i(\sum_j|A_{ij}|^2)(\sum_j|x_j|^2)=\langle A\rangle_*\|x\|_2^2$ by Cauchy–Schwarz. For the second, $\langle A\rangle_*=\sum_i\|Ae_i\|_2^2\le\sum_i\|A\|^2=5\|A\|^2$.
Corollary (Hadamard). $|\det A|\le\|A\|^5$.
Proof. The determinant is the product of the norms of the columns times the volume of the parallelepiped, which is at most $1$; more directly, Hadamard's inequality gives $|\det A|\le\prod_i\|a_i\|_2\le\|A\|^5$, because each column has norm at most the operator norm.
Remark. The operator norm is not polynomial. It is the spectral norm, equal to the square root of the largest eigenvalue of $A^*A$; the spectral theorem for $A^*A$ is Spectral Theory and the Spectral Projectors of $M_5(\mathbb{C})$. The inequality with the Frobenius norm is proved here without the spectral theorem, and it is the only comparison needed in this article.
The Singular Set
The Determinant Hypersurface
Definition. The singular set is
$$ \Sigma=\{A\in M_5(\mathbb{C}):\det A=0\}. $$
Theorem. $\Sigma$ is a complex algebraic hypersurface of complex dimension $24$ and real dimension $48$; its complement $GL_5(\mathbb{C})$ is a nonempty Zariski open set of complex dimension $25$ and real dimension $50$.
Proof. The determinant is a nonconstant polynomial on $M_5\cong\mathbb{C}^{25}$; its zero locus is a hypersurface of complex dimension $24$; the complement is the nonvanishing locus of a polynomial, a nonempty Zariski open set. The real dimensions are twice the complex dimensions.
Corollary. $\Sigma$ has real codimension $2$. At a point of rank $4$, the derivative of the determinant is nonzero, and the zero set is locally the zero set of one complex equation, equivalently two independent real equations.
Proof. At rank $4$, $\operatorname{adj}(A)\ne0$, so the derivative $H\mapsto\operatorname{tr}(\operatorname{adj}(A)H)$ is a nonzero complex-linear functional; its kernel has real codimension $2$.
The Rank Strata
Definition. The determinantal variety of rank at most $r$ is
$$ M_5^{(r)}=\{A:\operatorname{rank}A\le r\},\qquad r=0,1,\dots,5 . $$
Theorem. $M_5^{(r)}$ is an irreducible algebraic variety of complex dimension
$$ \dim M_5^{(r)}=r(10-r), $$
and complex codimension $(5-r)^2$. The exact rank-$r$ stratum $M_5^{(r)}\setminus M_5^{(r-1)}$ has the same dimension $r(10-r)$ and is smooth.
Proof. A rank-$r$ matrix factors as $A=UV$ with $U\in\mathbb{C}^{5\times r}$ and $V\in\mathbb{C}^{r\times5}$; the pair $(U,V)$ has $10r$ parameters, and the ambiguity is the action of $GL_r$, of dimension $r^2$; hence the dimension is $10r-r^2=r(10-r)$. The exact rank-$r$ matrices form a homogeneous space of $GL_5\times GL_5$ and are smooth; the lower-rank loci are the singular strata. The codimension is $25-r(10-r)=(5-r)^2$.
Corollary. The dimensions and codimensions are
| $r$ | $\dim M_5^{(r)}$ | complex codimension | real codimension |
|---|---|---|---|
| $0$ | $0$ | $25$ | $50$ |
| $1$ | $9$ | $16$ | $32$ |
| $2$ | $16$ | $9$ | $18$ |
| $3$ | $21$ | $4$ | $8$ |
| $4$ | $24$ | $1$ | $2$ |
| $5$ | $25$ | $0$ | $0$ |
Proof. Evaluate $r(10-r)$ and $(5-r)^2$; the real codimension is twice the complex codimension.
Corollary. The singular set is $\Sigma=M_5^{(4)}$, of complex dimension $24$ and real codimension $2$. Its singular locus is $M_5^{(3)}$, of complex dimension $21$; the successive singular loci are the lower-rank varieties.
Proof. The derivative of the determinant vanishes exactly where the adjugate vanishes, i.e., at rank at most $3$; the rank-$4$ points are smooth. The singular locus of $M_5^{(3)}$ is $M_5^{(2)}$, and so on.
Remark. The successive codimensions of the rank strata are the odd numbers $1,3,5,7,9$. The codimensions are the squares $1,4,9,16,25$; their differences are $3,5,7,9$ and the first is $1$. These are the dimensions of the adjoint summands $V_0,\dots,V_4$ of the principal $\mathfrak{sl}_2$. The coincidence is a numerical pattern of the two stratifications; the rank strata and the adjoint summands are different objects.
The Singular Locus
Proposition. The singular locus of $\Sigma$ is $M_5^{(3)}$, the matrices of rank at most $3$. The smooth locus of $\Sigma$ is the exact rank-$4$ stratum.
Proof. The gradient of the determinant is the adjugate; the adjugate vanishes exactly when the rank is at most $3$ (all $4\times4$ minors vanish). At rank $4$, at least one $4\times4$ minor is nonzero, so the gradient is nonzero and the hypersurface is smooth.
Corollary. The rank stratification is a Whitney stratification of $\Sigma$; the codimensions in $\Sigma$ of the strata of ranks $3,2,1,0$ are $3,8,15,24$, and the successive differences are $3,5,7,9$.
Proof. The codimension in $\Sigma$ of $M_5^{(r)}$ is $(5-r)^2-1$ for $r\le3$: for $r=3$, $4-1=3$; for $r=2$, $9-1=8$; for $r=1$, $16-1=15$; for $r=0$, $25-1=24$. The successive differences are $5,7,9$ after the initial $3$. The determinantal varieties are classic examples of Whitney stratifications.
The Norm and the Singular Set
No Null Cone
Theorem. The Frobenius form is positive definite, so it has no null cone; the only null vector is zero. The singular set $\Sigma$ is not a null cone of the Frobenius form.
Proof. $\langle A,A\rangle_F=0$ implies $A=0$; $\Sigma$ contains nonzero matrices (the nilpotents), so it is not the null set.
Corollary. The determinant zero set and the isotropic cone of the trace pairing are different hypersurfaces. The first is the set of non-invertibles; the second is the set of matrices with $\operatorname{tr}(A^2)=0$.
Proof. The determinant vanishes on every singular matrix; $\operatorname{tr}(A^2)$ vanishes on the nilpotents and on many other matrices; the two equations are different.
The Norm's Zero Set
Proposition. The zero set of the determinant norm is $\Sigma$; the norm is not positive, and its zero set is the singular set. The Frobenius norm and the operator norm have zero set $\{0\}$.
Proof. $\det A=0$ iff $A\in\Sigma$; the analytic norms vanish only at the zero matrix.
Corollary. The determinant norm is a multiplicative form with zeros, not a norm; the singular set is its zero locus. The two analytic norms are positive definite and have no zeros except $0$; they define the Hilbert space structure of The Involution and the Hermitian Structure of $M_5(\mathbb{C})$.
Proof. The multiplicativity and the zeros are the determinant properties; the positivity of the analytic norms is the sum of squares.
Remark. The singular set is $\det=0$, of real codimension $2$, and it is not a null cone. The determinant norm is a multiplicative form of degree $5$; its zero set is the singular set. The positive definite forms have no null cone; the indefinite trace pairing has an isotropic cone, but that cone is a different object. The three notions — zeros of the determinant, isotropic vectors of the trace pairing, null vectors of the Frobenius form — must be kept apart.
Summary
The determinant is a multiplicative form of degree $5$; it is the norm of the algebra, but it is not a norm in the analytic sense. The analytic norms are the Frobenius norm and the operator norm, related by $\|A\|\le\langle A\rangle_*^{1/2}\le\sqrt5\|A\|$. The determinant is bounded by the operator norm, $|\det A|\le\|A\|^5$ (Hadamard).
$$ \boxed{\ \det\text{ is a multiplicative form of degree }5;\quad \Sigma=\{\det=0\}\text{ has real codimension }2;\quad \text{no null cone.}\ } $$
The singular set is the determinantal hypersurface of rank at most $4$, of complex dimension $24$ and real codimension $2$. The rank strata have complex dimensions $0,9,16,21,24$ and complex codimensions $25,16,9,4,1$; the successive codimensional differences are $9,7,5,3$, and the codimension of $\Sigma$ is $1$, so the odd numbers in reverse order are $9,7,5,3,1$. The singular locus of $\Sigma$ is the rank-at-most-$3$ variety. The Frobenius form is positive definite and has no null cone; the trace pairing is indefinite and has an isotropic cone; the determinant zero set is neither.
| Object | Value |
|---|---|
| Determinant | Multiplicative form of degree $5$ |
| Frobenius norm | $\sqrt{\sum_{ij}|A_{ij}|^2}$ |
| Operator norm | $\sup_{\|x\|=1}\|Ax\|$ |
| Norm comparison | $\|A\|\le\langle A\rangle_*^{1/2}\le\sqrt5\|A\|$ |
| Hadamard | $|\det A|\le\|A\|^5$ |
| Singular set | $\det=0$, complex dimension $24$ |
| Real codimension | $2$ |
| Rank strata | dimensions $0,9,16,21,24$ |
| Codimensions | $25,16,9,4,1$ |
| Null cone | none for the Frobenius form |
Summary of Notation
| Symbol | Meaning |
|---|---|
| $N(A)=\det A$ | The determinant norm |
| $\langle A\rangle_*^{1/2}$ | The Frobenius norm |
| $\|A\|$ | The operator norm |
| $\langle A,B\rangle_{\mathrm{tr}}=\operatorname{tr}(AB)$ | The trace pairing, indefinite |
| $\langle A,B\rangle_F=\operatorname{tr}(A^*B)$ | The Frobenius form, positive definite |
| $\Sigma=\{\det=0\}$ | The singular set |
| $M_5^{(r)}$ | The matrices of rank at most $r$ |
Further Reading
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge University Press, second edition, 2013), for the determinant, the operator norm, Hadamard's inequality and the singular values.
- Israel M. Gelfand, Mikhail M. Kapranov and Andrei V. Zelevinsky, Discriminants, Resultants, and Multidimensional Determinants (Birkhäuser, 1994), for the determinant as a form, the discriminants and the determinantal varieties.
- Joe Harris, Algebraic Geometry: A First Course (Springer, 1992), for the determinantal varieties, their dimensions and their singular loci.
- The Involution and the Hermitian Structure of $M_5(\mathbb{C})$, for the positive definite Frobenius form and the real codimension $2$.
- The Element, the Coordinates and the Trace of $M_5(\mathbb{C})$, for the trace pairing and the determinant.
- The Quintic and the Spectral Strata of $M_5(\mathbb{C})$, for the determinant, the discriminant and the spectral strata.
- Spectral Theory and the Spectral Projectors of $M_5(\mathbb{C})$, for the spectral theorem and the operator norm as the largest singular value.