The Two Base Fields: $M_4(\mathbb{R})$ inside $M_4(\mathbb{C})$
Introduction
The real matrices are the fixed set of the entrywise conjugation of $M_4(\mathbb{C})$, a real form of real dimension $16$ inside $32$, and inside them sit the two ladders $\mathbb{R}\subset\mathbb{C}$ and $\mathbb{R}\subset\mathbb{H}$, the first with a central imaginary unit and the second with three anticommuting ones.
The article proves that $M_4(\mathbb{R})$ is the fixed algebra of $\tau(A)=\bar A$, that $\mathbb{C}\otimes_{\mathbb{R}}M_4(\mathbb{R})\cong M_4(\mathbb{C})$, that the rank survives the change of field while every dimension doubles, and that in the axis-free basis the transpose is the quaternion conjugation. It closes with what fails over $\mathbb{R}$: the exponential is not onto, the group has two components, and the spectrum leaves the base field.
It assumes The Matrix Algebra $M_4(\mathbb{C})$ for the product, the trace and the rank, and $M_4(\mathbb{C})$ Axis-Free Representation for the coefficient basis. The general theory of the pair $M_n(\mathbb{R})\subset M_n(\mathbb{C})$ is The Two Base Fields: $M_n(\mathbb{R})$ inside $M_n(\mathbb{C})$, and the second real form of the complex algebra is The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$.
The Real Matrices as a Fixed Set
Definition. The entrywise conjugation of $M_4(\mathbb{C})$ is $\tau(A)=\bar A$, applied to the sixteen entries.
Proposition. The map $\tau$ is an involutive antilinear automorphism: it is additive, $\tau(\lambda A)=\bar\lambda\tau(A)$, multiplicative, $\tau(AB)=\tau(A)\tau(B)$, and $\tau^{2}=\mathrm{id}$.
Proof. Conjugation of complex numbers is additive, conjugate-linear and multiplicative, and the matrix product is a sum of products of entries; hence $\overline{AB}=\bar A\bar B$. Two conjugations return each entry.
Theorem. The fixed set of $\tau$ is $M_4(\mathbb{R})$, a real subalgebra of real dimension $16$.
Proof. An entry is fixed exactly when it is real. The real matrices are closed under sums, products and real scalars, so they form a real subalgebra; the real dimension is the count of the sixteen entries.
The fixed set is not a complex subspace and not an ideal. It is stable under real scalars only, since $i\mathrm I_4$ is not real, and the product $E_{11}(iE_{11})=iE_{11}$ leaves it. It is a real form, because $M_4(\mathbb{R})\cap iM_4(\mathbb{R})=0$ and the two sum to the algebra.
The Complexification and the Doubling of Dimension
Theorem. $\mathbb{C}\otimes_{\mathbb{R}}M_4(\mathbb{R})\cong M_4(\mathbb{C})$ as $\mathbb{C}$-algebras.
Proof. The map $\lambda\otimes X\mapsto\lambda X$ is $\mathbb{C}$-bilinear and multiplicative, since $(\lambda\otimes X)(\mu\otimes Y)=\lambda\mu\otimes XY$. It is surjective because $A=X+iY$ with $X=\tfrac12(A+\bar A)$ and $Y=\tfrac1{2i}(A-\bar A)$ real, and the two sides have complex dimension $16$.
Corollary (the real and imaginary parts). Every matrix has a unique decomposition
$$ A=X+iY,\qquad X=\tfrac12(A+\bar A)\in M_4(\mathbb{R}),\quad Y=\tfrac1{2i}(A-\bar A)\in M_4(\mathbb{R}), $$
and the product reads $AC=(XU-YV)+i(XV+YU)$ for $C=U+iV$ with $U,V$ real.
The dimension count. The real dimension of $M_4(\mathbb{C})$ is $32$; that of $M_4(\mathbb{R})$ is $16$. The groups follow: $\dim_{\mathbb{R}}\mathrm{GL}_4(\mathbb{C})=32$ and $\dim_{\mathbb{R}}\mathrm{GL}_4(\mathbb{R})=16$.
Proposition (the rank survives). For $X\in M_4(\mathbb{R})$ the rank over $\mathbb{R}$ equals the rank over $\mathbb{C}$.
Proof. The rank is the largest order of a nonvanishing minor, and a minor is a polynomial with integer coefficients in the entries. Its vanishing is a condition on the entries and not on the field in which they are read.
Corollary. The rank strata meet $M_4(\mathbb{R})$ in real dimensions $0$, $7$, $12$, $15$ and $16$, since a rank-$k$ stratum has dimension $8k-k^{2}$ and the real slice has the same real dimension.
The Two Ladders
The complex ladder. Inside $M_4(\mathbb{R})$ sits the scalar line $\mathbb{R}\mathrm I_4$; adjoining one matrix $J$ with $J^{2}=-\mathrm I_4$ gives a copy of $\mathbb{C}$, of real dimension $2$. The three levels are $\mathbb{R}\subset\mathbb{C}\subset M_4(\mathbb{R})$, of real dimensions $1,2,16$.
The quaternionic ladder. The real quaternions sit inside $M_4(\mathbb{R})$ as the left multiplication $q\mapsto L_{\mathrm{axf}}(q)$ with real coefficients,
$$ q=q_te_t+q_x\vec e_x+q_y\vec e_y+q_z\vec e_z,\qquad q_\mu\in\mathbb{R}. $$
The four matrices $L_{\mathrm{axf},e_\mu}$ are the real signed permutations of $M_4(\mathbb{C})$ Axis-Free Representation. They obey $L_{\mathrm{axf},\vec e_x}^{2}=L_{\mathrm{axf},\vec e_y}^{2}=L_{\mathrm{axf},\vec e_z}^{2}=-\mathrm I_4$ and the cyclic product rule, so the image is a copy of $\mathbb{H}$ of real dimension $4$.
Proposition. The complex ladder sits inside the quaternionic one: $\mathbb{R}\subset\mathbb{C}\subset\mathbb{H}\subset M_4(\mathbb{R})$, of real dimensions $1,2,4,16$.
Proof. The scalars are $q=q_te_t$; the copy of $\mathbb{C}$ is spanned by $e_t$ and one vector unit, which squares to $-e_t$ and commutes with itself; the three vector units together span $\mathbb{H}$. The inclusions are strict by the dimensions.
The two ladders differ in the centre. In $\mathbb{C}$ the imaginary unit is central, and $i\mathrm I_4$ is central in $M_4(\mathbb{C})$; in $\mathbb{H}$ no vector unit is central, since $\vec e_x\vec e_y=\vec e_z=-\vec e_y\vec e_x$. The complex ladder is the scalar field of the algebra, the quaternionic ladder a division subalgebra of it: $\langle q\rangle_{\natural}=q_t^{2}+q_x^{2}+q_y^{2}+q_z^{2}$ is a sum of four real squares, so $\det L_{\mathrm{axf}}(q)=\langle q\rangle_{\natural}^{2}$ vanishes only at $q=0$.
The Real Slice of the Axis-Free Frame
The sixteen matrices are real. Each $T_{\mu\nu}=L_{\mathrm{axf},e_\mu}R_{\mathrm{axf},e_\nu}$ is a signed permutation matrix, hence real, and the sixteen are independent over $\mathbb{C}$. Their real span is a subspace of $M_4(\mathbb{R})$ of real dimension $16$, which is all of it.
Proposition. No proper real subspace contains the sixteen, and the expansion $A=\sum q_{\mu\nu}T_{\mu\nu}$ of a real matrix has real coefficients $q_{\mu\nu}=\tfrac14\operatorname{tr}(T_{\mu\nu}^{\mathsf T}A)$.
Proof. Sixteen independent real matrices in a real space of dimension sixteen span it, and each trace is real for real $A$; the formula is the orthogonality $\operatorname{tr}(T_{\mu\nu}^{\mathsf T}T_{\rho\sigma})=4\delta_{\mu\rho}\delta_{\nu\sigma}$ read as a projection.
The trace pairing on the slice. In this basis the pairing is diagonal,
$$ \operatorname{tr}(T_{\mu\nu}T_{\rho\sigma})=\pm4\,\delta_{\mu\rho}\delta_{\nu\sigma}, $$
with $+4$ on the ten symmetric elements and $-4$ on the six antisymmetric ones. The inertia $(10,6)$ is therefore a statement about the real algebra already, and the complex algebra inherits it in the same basis. The full account is The Trace Form and Its Inertia on the Two Halves of $M_4(\mathbb{C})$.
One count is not the other. The slice of the algebra itself is four-dimensional, the image of $\mathbb{H}$; the sixteen is the dimension of the operator basis, in which elements of the algebra appear as vectors.
The Transpose in the Real Slice
The transpose is the quaternion conjugation. In the axis-free frame,
$$ L_{\mathrm{axf}}(\tilde Q)^{\mathsf T}=L_{\mathrm{axf}}(\tilde Q^{\natural}),\qquad \tilde Q^{\natural}=Q_te_t-Q_x\vec e_x-Q_y\vec e_y-Q_z\vec e_z . $$
On the real slice this is a real identity: the transpose of a real matrix is real, and it is the matrix of the element with the three vector coefficients negated.
Proposition. The identity holds on real coefficients, and the transpose is an antiautomorphism of the image $\mathbb{B}_L$.
Proof. Transposition reverses products, $(AB)^{\mathsf T}=B^{\mathsf T}A^{\mathsf T}$, and the quaternion conjugation does the same, $(q_1q_2)^{\natural}=q_2^{\natural}q_1^{\natural}$. The identity on the four units extends by linearity.
The dagger on the slice. The conjugate transpose of a real matrix is its transpose, so ${}^{\ast}$ restricts to ${}^{\mathsf T}$ on $M_4(\mathbb{R})$. Hence $\operatorname{Herm}_4\cap M_4(\mathbb{R})=\operatorname{Sym}_4(\mathbb{R})$, of real dimension $10$, and $\mathfrak{u}(4)\cap M_4(\mathbb{R})=\operatorname{Skew}_4(\mathbb{R})$, of real dimension $6$: the complex decomposition of dimensions $16+16$ meets the real slice in the decomposition $10+6$.
What Fails over $\mathbb{R}$
The exponential is not onto. Over $\mathbb{C}$ the exponential is surjective onto $\mathrm{GL}_4(\mathbb{C})$, by the Jordan form and a logarithm chosen on each eigenvalue. Over $\mathbb{R}$ it is not. If $e^{X}=A$ with $X$ real, the eigenvalues of $X$ are logarithms of those of $A$; at a one-dimensional block with the eigenvalue $-2$ the only logarithm is $\log2+i\pi$, and a real $X$ would need $\log2-i\pi$ as well, which is a logarithm of no eigenvalue of $A$. Hence $\operatorname{diag}(-1,-1,-2,-3)$ has no real logarithm.
The group has two components. $\mathrm{GL}_4(\mathbb{R})$ has two connected components, separated by the sign of the determinant, while $\mathrm{GL}_4(\mathbb{C})$ is connected. The exponential lands in the component of positive determinant, and even there not onto it.
The spectrum leaves the base field. A real matrix has a characteristic polynomial with real coefficients, so its nonreal eigenvalues occur in conjugate pairs and its Jordan form carries $2\times2$ rotation blocks. The block of angle $\theta$ has the eigenvalues $e^{\pm i\theta}$ and no real eigenvector unless $\theta\in\pi\mathbb{Z}$.
The quaternionic norm becomes Euclidean. Over $\mathbb{C}$ the form $\langle\tilde Q\rangle_{\natural}$ is isotropic: the element with $Q_t=1$ and $Q_x=i$ has norm $1-1=0$, and its left matrix is a nonzero zero divisor. On the real slice the same form is a sum of four real squares, and its zero set collapses to the origin. One formula, a Lorentzian form in one reading and a Euclidean norm in the other.
Summary
$M_4(\mathbb{R})$ is the fixed algebra of the entrywise conjugation of $M_4(\mathbb{C})$, of real dimension $16$ against $32$, and $\mathbb{C}\otimes_{\mathbb{R}}M_4(\mathbb{R})\cong M_4(\mathbb{C})$. Every matrix is $X+iY$ with $X,Y$ real and unique, the rank is a condition on minors and survives the change of field, and the rank strata have real dimensions $0,7,12,15,16$.
Two ladders sit inside the real matrices: $\mathbb{R}\subset\mathbb{C}\subset M_4(\mathbb{R})$ with a central imaginary unit, and $\mathbb{R}\subset\mathbb{C}\subset\mathbb{H}\subset M_4(\mathbb{R})$ with three anticommuting ones. The second is a division algebra of real dimension $4$, and it is the real slice of the biquaternion algebra under the left regular representation.
The axis-free basis is real, so its sixteen matrices are a coordinate frame of $M_4(\mathbb{R})$ and the trace pairing is diagonal there with ten signs $+4$ and six signs $-4$. On the real slice the transpose is the quaternion conjugation, and the dagger restricts to the transpose. What fails over $\mathbb{R}$ is the surjectivity of the exponential, the connectedness of the group, and the closedness of the spectrum; and the quaternionic norm turns from a Lorentzian form into a Euclidean one.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\tau(A)=\bar A$ | the entrywise conjugation, an antilinear involution |
| $M_4(\mathbb{R})$ | the fixed set of $\tau$, of real dimension $16$ |
| $X,Y$ | the real and imaginary parts of $A=X+iY$ |
| $L_{\mathrm{axf}}$ | the left regular representation, real on real coefficients |
| $T_{\mu\nu}$ | the sixteen two-sided basis matrices, all real |
| $q_{\mu\nu}$ | the real coordinates of a real matrix in that basis |
| $\natural$ | the quaternion conjugation, equal to the transpose in this frame |
| $\langle\tilde Q\rangle_{\natural}$ | $Q_t^{2}+Q_x^{2}+Q_y^{2}+Q_z^{2}$ |
Further Reading
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge University Press, 2013), for the real and complex readings of one matrix and the reality of the spectrum.
- The Matrix Algebra $M_4(\mathbb{C})$, for the product, the trace, the determinant and the rank of the complex algebra.
- $M_4(\mathbb{C})$ Axis-Free Representation, for the coefficient basis, the four left matrices and the transpose identity.
- The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$, for the second real form and its classification.
- The Two Base Fields: $M_2(\mathbb{R})$ inside $M_2(\mathbb{C})$, for the same pair at order two.