The Symplectic Group Sp(2n, R) and Its Role in Linear Canonical Transformations

Intuition Beginner

Unit 09.05.01 introduced canonical transformations as relabellings of phase space that preserve Hamilton's equations. Unit 09.05.03 showed how to construct them using generating functions. Now we ask: what happens when the transformation is linear -- when the new coordinates are linear combinations of the old ones?

For a single degree of freedom, a linear transformation of the $(q,p)$ plane is specified by a $2\times 2$ matrix:

$$\left(\begin{array}{c}Q\\ P\end{array}\right)=M\left(\begin{array}{c}q\\ p\end{array}\right),M=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right).$$

Not every such matrix produces a canonical transformation. The requirement that Hamilton's equations survive imposes a constraint: $\det M=ad-bc=1$. For one degree of freedom, the symplectic group $\mathrm{Sp}(2,\mathbb{R})$ is exactly the group of $2\times 2$ real matrices with determinant 1, also known as $\mathrm{SL}(2,\mathbb{R})$.

The determinant-1 condition has a geometric meaning: area preservation. A matrix with $\det M=1$ takes any region of the $(q,p)$ plane and maps it to a region with the same area. This is Liouville's theorem at the linear level -- phase-space area is conserved under canonical transformations. For $n>1$ degrees of freedom, the symplectic group is a proper subgroup of the volume-preserving matrices: it preserves not just the $2n$-dimensional volume but the specific pairing between each position $q_i$ and its conjugate momentum $p_i$.

Three families of $2\times 2$ symplectic matrices appear throughout physics:

Rotations. $R(\theta )=\left(\begin{array}{cc}\cos \theta & \sin \theta \\ -\sin \theta & \cos \theta \end{array}\right)$. These rotate the phase plane by angle $\theta$. The harmonic oscillator flow $\dot{q}=p,\dot{p}=-{\omega }^{2}q$ traces out circles in phase space; the time-evolution matrix is $R(\omega t)$, a one-parameter family of rotations.

Squeezes. $S(\lambda )=\left(\begin{array}{cc}\lambda & 0\\ 0& 1/\lambda \end{array}\right)$ for $\lambda >0$. These stretch the $q$-direction by $\lambda$ and compress the $p$-direction by $1/\lambda$. But wait -- is this symplectic? We need $\det S=1$. And indeed $\lambda \cdot (1/\lambda )=1$, so $S(\lambda )\in \mathrm{Sp}(2,\mathbb{R})$ for any $\lambda >0$. The squeeze preserves area while distorting shape: an ellipse squeezed in one direction bulges in the other. In quantum optics, squeezed states of light are produced by Hamiltonians of the form $H=\frac{1}{2}(q^2+p^2)+ϵqp$.

Wait -- the matrix $S(\lambda )=\left(\begin{array}{cc}\lambda & 0\\ 0& 1/\lambda \end{array}\right)$ has $\det S=1$, but is it symplectic in the sense of preserving the $q$-$p$ pairing? Let us verify directly. For $n=1$ the symplectic form is $J=\left(\begin{array}{cc}0& 1\\ -1& 0\end{array}\right)$. Compute:

$S^{T}JS=\left(\begin{array}{cc}\lambda & 0\\ 0& 1/\lambda \end{array}\right)\left(\begin{array}{cc}0& 1\\ -1& 0\end{array}\right)\left(\begin{array}{cc}\lambda & 0\\ 0& 1/\lambda \end{array}\right)=\left(\begin{array}{cc}0& \lambda \\ -1/\lambda & 0\end{array}\right)\left(\begin{array}{cc}\lambda & 0\\ 0& 1/\lambda \end{array}\right)=\left(\begin{array}{cc}0& 1\\ -1& 0\end{array}\right)=J.$

Yes, the squeeze is symplectic. Area preservation alone is not sufficient in general (for $n>1$ it fails), but for $n=1$ the conditions coincide.

Shears. $T(\alpha )=\left(\begin{array}{cc}1& \alpha \\ 0& 1\end{array}\right)$ for real $\alpha$. This shifts $Q=q+\alpha p$ while leaving $P=p$ unchanged. A shear slides the phase plane parallel to the $q$-axis by an amount proportional to $p$. Its determinant is 1, so it is symplectic. The generating function for this transformation (unit 09.05.03) is $F_2(q,P)=qP+\frac{\alpha }{2}{P}^2$.

Any $2\times 2$ symplectic matrix can be decomposed as a product of rotations, squeezes, and shears. This is the analogue of the polar decomposition for general invertible matrices, specialised to the symplectic setting.

Visual Beginner

The three fundamental transformations of $\mathrm{Sp}(2,\mathbb{R})$ acting on the unit circle in the $(q,p)$ plane:

Transformation	Matrix	Effect on unit circle
Rotation $R(\theta )$	$\left(\begin{array}{cc}\cos \theta & \sin \theta \\ -\sin \theta & \cos \theta \end{array}\right)$	Circle stays a circle; rotates by $\theta$
Squeeze $S(\lambda )$	$\left(\begin{array}{cc}\lambda & 0\\ 0& 1/\lambda \end{array}\right)$	Circle becomes ellipse; area preserved
Shear $T(\alpha )$	$\left(\begin{array}{cc}1& \alpha \\ 0& 1\end{array}\right)$	Circle becomes ellipse; tilted axis

Every element of $\mathrm{Sp}(2,\mathbb{R})$ deforms the unit circle into an ellipse of the same area. The rotation determines the tilt of the ellipse, the squeeze determines the eccentricity, and the shear introduces an asymmetry. Conversely, every area-preserving ellipse can be reached by a unique combination of these three operations.

Worked example Beginner

Verifying symplecticity of a general $2\times 2$ matrix

Let $M=\left(\begin{array}{cc}3& 2\\ 1& 1\end{array}\right)$. Is it symplectic?

Compute $\det M=3\cdot 1-2\cdot 1=1$. For $n=1$, this is both necessary and sufficient: $M\in \mathrm{Sp}(2,\mathbb{R})$.

Verify the symplectic condition directly. $M^{T}JM=\left(\begin{array}{cc}3& 1\\ 2& 1\end{array}\right)\left(\begin{array}{cc}0& 1\\ -1& 0\end{array}\right)\left(\begin{array}{cc}3& 2\\ 1& 1\end{array}\right)=\left(\begin{array}{cc}-1& 3\\ -1& 2\end{array}\right)\left(\begin{array}{cc}3& 2\\ 1& 1\end{array}\right)=\left(\begin{array}{cc}0& 1\\ -1& 0\end{array}\right)=J.$ Confirmed.

The transformation is $Q=3q+2p$, $P=q+p$. This is neither a pure rotation, squeeze, nor shear -- it is a composition of all three.

The harmonic oscillator as a rotation

The harmonic oscillator Hamiltonian $H=\frac{1}{2}(p^2+{\omega }^2{q}^2)$ gives equations of motion $\dot{q}=p$, $\dot{p}=-{\omega }^{2}q$. Rescaling $q\to q$ and $p\to p/\omega$, the equations become $\dot{q}=\omega \tilde{p}$, $\dot{\tilde{p}}=-\omega q$, whose solution is the rotation:

$$\left(\begin{array}{c}q(t)\\ \tilde{p}(t)\end{array}\right)=\left(\begin{array}{cc}\cos \omega t& \sin \omega t\\ -\sin \omega t& \cos \omega t\end{array}\right)\left(\begin{array}{c}q(0)\\ \tilde{p}(0)\end{array}\right).$$

The time-evolution matrix is a rotation by angle $\omega t$. Since rotations are symplectic, the flow preserves phase-space area -- a manifestation of Liouville's theorem for the harmonic oscillator.

A squeeze that is area-preserving but not symplectic (for $n>1$)

For two degrees of freedom, consider $M=\left(\begin{array}{cccc}2& 0& 0& 0\\ 0& 1/2& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)$. This stretches $q_1$ by 2 and compresses $p_1$ by $1/2$, leaving $(q_2,p_2)$ unchanged. The determinant is $2\cdot (1/2)\cdot 1\cdot 1=1$: volume is preserved.

But compute $M^{T}JM$ where $J=\left(\begin{array}{cc}0& I_2\\ -I_2& 0\end{array}\right)$:

$M^{T}JM=\left(\begin{array}{cccc}0& 0& 0& 2\\ 0& 0& -1/2& 0\\ 0& 1/2& 0& 0\\ -2& 0& 0& 0\end{array}\right).$

This is not $J$: the $(1,4)$ and $(4,1)$ entries are $2$ and $-2$ instead of $1$ and $-1$. The matrix preserves volume but scrambles the $q_1$-$p_1$ pairing. For $n>1$, symplecticity is strictly stronger than volume preservation.

Check your understanding Beginner

Formal definition Intermediate+

The symplectic group

Let $V$ be a $2n$-dimensional real vector space equipped with the standard symplectic form $\omega$, represented in coordinates by the matrix $J=\left(\begin{array}{cc}0& I_n\\ -I_n& 0\end{array}\right)$.

Definition. A $2n\times 2n$ real matrix $M$ is symplectic if

$$\boxed{M^{T}JM=J.}$$

The set of all symplectic matrices forms the symplectic group $\mathrm{Sp}(2n,\mathbb{R})$ under matrix multiplication.

The condition $M^{T}JM=J$ says that $M$ preserves the symplectic pairing: $\omega (Mu,Mv)=\omega (u,v)$ for all vectors $u,v$.

Basic properties

Group properties. $\mathrm{Sp}(2n,\mathbb{R})$ is closed under:

• Multiplication. If $M_1,M_2\in \mathrm{Sp}(2n,\mathbb{R})$, then $(M_1{M}_2{)}^{T}J(M_1{M}_2)=M_2^T{M}_1^{T}J{M}_1{M}_2=M_2^{T}J{M}_2=J$.
• Inverse. From $M^{T}JM=J$, rearranging gives $M^{-1}=J^{-1}{M}^{-T}{J}^T=-J{M}^{T}J$ (using $J^{-1}=-J$). Then $(M^{-1}{)}^{T}J{M}^{-1}=(-J{M}^{T}J{)}^{T}J(-J{M}^{T}J)=J^{T}M{J}^{T}J(-J{M}^{T}J)=J$ after simplification.
• Identity. $I^{T}JI=J$.

Determinant. Taking $\det$ of both sides of $M^{T}JM=J$: $(\det M{)}^2\det J=\det J$. Since $\det J=1$, this gives $(\det M{)}^2=1$, so $\det M=\pm 1$. Because $\mathrm{Sp}(2n,\mathbb{R})$ is connected and contains the identity ($\det I=1$), every symplectic matrix has $\det M=+1$.

Block structure

Write $M=\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)$ with $n\times n$ blocks. The condition $M^{T}JM=J$ expands to three independent equations:

$$A^{T}C=C^{T}A,B^{T}D=D^{T}B,A^{T}D-C^{T}B=I_n.$$

The first two say that $A^{T}C$ and $B^{T}D$ are symmetric. The third is the generalised area-preservation condition. For $n=1$ these reduce to the single equation $ad-bc=1$.

An equivalent set of conditions is obtained from $MJ{M}^T=J$:

$$A{B}^T=B{A}^T,C{D}^T=D{C}^T,A{D}^T-B{C}^T=I_n.$$

The Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$

The tangent space at the identity of $\mathrm{Sp}(2n,\mathbb{R})$ is the symplectic Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$, consisting of matrices $X$ satisfying:

$$X^{T}J+JX=0\text{(equivalently, }JX\text{ is symmetric).}$$

In block form $X=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)$, the conditions are $b=b^T$, $c=c^T$, $d=-a^T$. The dimension is $n(2n+1)$.

The Lie bracket is the matrix commutator $[X,Y]=XY-YX$. The exponential map $\exp :\mathfrak{sp}(2n,\mathbb{R})\to \mathrm{Sp}(2n,\mathbb{R})$ sends every Lie-algebra element to a symplectic matrix, though it is not surjective for $n\ge 2$.

One-parameter subgroups: rotations, squeezes, and shears

The generators of $\mathrm{Sp}(2n,\mathbb{R})$ are best understood through one-parameter subgroups $\exp (tX)$ for $X\in \mathfrak{sp}(2n,\mathbb{R})$:

Rotations (elliptic generators). For a single pair $(q_i,p_i)$, $X=\left(\begin{array}{cc}0& e_i{e}_i^T\\ -e_i{e}_i^T& 0\end{array}\right)$ generates $R_i(\theta )=\exp (\theta X)$, a rotation in the $(q_i,p_i)$ plane. These have eigenvalues on the unit circle.

Squeezes (hyperbolic generators). $X=\left(\begin{array}{cc}\sigma & 0\\ 0& -\sigma \end{array}\right)$ with $\sigma$ diagonal generates $S({\lambda }_1,\dots ,{\lambda }_n)=\exp (tX)$, where ${\lambda }_i=e^{t{\sigma }_i}$. Each pair $(q_i,p_i)$ is independently squeezed.

Shears (parabolic generators). $X=\left(\begin{array}{cc}0& S\\ 0& 0\end{array}\right)$ with $S$ symmetric generates $T=\exp (tX)=\left(\begin{array}{cc}I& tS\\ 0& I\end{array}\right)$, a shear in the position coordinates.

The Iwasawa decomposition

Every $M\in \mathrm{Sp}(2n,\mathbb{R})$ can be uniquely decomposed as $M=K\cdot A\cdot N$ where:

• $K=\mathrm{Sp}(2n,\mathbb{R})\cap \mathrm{O}(2n)\cong \mathrm{U}(n)$ is the maximal compact subgroup (the rotations),
• $A$ is the group of positive-diagonal squeezes,
• $N$ is the group of upper-triangular shears with ones on the diagonal.

This is the Iwasawa decomposition of $\mathrm{Sp}(2n,\mathbb{R})$, analogous to the QR decomposition for general matrices. It provides a constructive parametrisation: any symplectic matrix is a rotation followed by a squeeze followed by a shear. For $n=1$, this says every $2\times 2$ matrix with $\det =1$ is a product $R(\theta )\cdot S(\lambda )\cdot T(\alpha )$, giving three real parameters matching the dimension $\dim \mathrm{Sp}(2,\mathbb{R})=3$.

Williamson's theorem

Theorem (Williamson, 1936). For any positive-definite symmetric $2n\times 2n$ matrix $S$, there exists $M\in \mathrm{Sp}(2n,\mathbb{R})$ such that

$$M^{T}SM=\left(\begin{array}{cc}\Lambda & 0\\ 0& \Lambda \end{array}\right),\Lambda =\mathrm{diag}({\lambda }_1,\dots ,{\lambda }_n),{\lambda }_i>0.$$

The numbers ${\lambda }_1,\dots ,{\lambda }_n$ are the symplectic eigenvalues of $S$ and are the positive eigenvalues of $JS$ (or equivalently, the positive square roots of the eigenvalues of $-JSJS$). They are invariant under symplectic conjugation.

Williamson's theorem says that any positive-definite quadratic form can be symplectically diagonalised. The symplectic eigenvalues play the role for symplectic geometry that ordinary eigenvalues play for orthogonal geometry. In classical mechanics, this theorem underlies the normal-mode decomposition of small oscillations 09.02.04: the quadratic Hamiltonian near a stable equilibrium can always be written as a sum of harmonic oscillators after a suitable linear canonical transformation.

In quantum optics, $S$ is the covariance matrix of a Gaussian state, and the symplectic eigenvalues are bounded below by $\hslash /2$ by the uncertainty principle.

Key derivation Intermediate+

Theorem. The eigenvalues of a symplectic matrix come in reciprocal pairs: if $\lambda$ is an eigenvalue, so is $1/\lambda$ (and $\bar{\lambda }$, $1/\bar{\lambda }$).

Proof. From $M^{T}JM=J$, rearrange to get $M^{-1}=-J{M}^{T}J$. Suppose $Mv=\lambda v$ with $v\ne 0$. Then $M^{-1}v={\lambda }^{-1}v$. But $M^{-1}=-J{M}^{T}J$, so $-J{M}^{T}Jv={\lambda }^{-1}v$, hence $M^T(Jv)=-{\lambda }^{-1}(Jv)$.

Now, $M$ and $M^T$ have the same eigenvalues (they have the same characteristic polynomial, since $\det (xI-M)=\det (xI-M^T)$). So if $-{\lambda }^{-1}$ is an eigenvalue of $M^T$, it is an eigenvalue of $M$.

But this gives $-{\lambda }^{-1}$, not $+{\lambda }^{-1}$. Where is the correction? The issue is the sign convention. Using $MJ{M}^T=J$ (which follows from $M^{T}JM=J$ and $\det M=1$), we get $M^{-T}=JM{J}^{-1}=-JMJ$. So $M^T=-J{M}^{-1}J$. If $Mv=\lambda v$, then $M^{-1}v={\lambda }^{-1}v$, so $M^T(-Jv)=-J{M}^{-1}v=-{\lambda }^{-1}(-Jv)={\lambda }^{-1}(Jv)$. Since $M$ and $M^T$ have the same eigenvalues, ${\lambda }^{-1}$ is an eigenvalue of $M$. Taking complex conjugates, ${\bar{\lambda }}^{-1}$ is also an eigenvalue.

The eigenvalues of a symplectic matrix therefore come in quartets $\{\lambda ,\bar{\lambda },{\lambda }^{-1},{\bar{\lambda }}^{-1}\}$, reducing to pairs $\{\lambda ,{\lambda }^{-1}\}$ when $\lambda$ is real, and pairs $\{e^{i\theta },e^{-i\theta }\}$ when $|\lambda |=1$. $\square$

Bridge. The eigenvalue structure of $\mathrm{Sp}(2n,\mathbb{R})$ constrains the linearised dynamics near any fixed point of a Hamiltonian flow. If the linearised map has an eigenvalue with $|\lambda |>1$ (the hyperbolic case), the equilibrium is unstable: nearby trajectories exponentially diverge. If all eigenvalues lie on the unit circle (the elliptic case), the equilibrium is linearly stable. The symplectic constraint means you cannot have a single eigenvalue leave the unit circle: eigenvalues leave in reciprocal pairs, and the transition from elliptic to hyperbolic passes through the collision point $\lambda =\pm 1$. This is the mechanism behind Hamiltonian bifurcations, and it is qualitatively different from the dissipative case where eigenvalues can cross the imaginary axis one at a time. The KAM theorem 09.08.01 describes the persistence of elliptic invariant tori under perturbation, and the eigenvalue constraints developed here are what give the Hamiltonian KAM theorem its specific structure.

Exercises Intermediate+

Lean formalization Intermediate+

Mathlib defines the symplectic group $\mathrm{Sp}(2n,\mathbb{R})$ in Mathlib.LinearAlgebra.SymplecticGroup as the set of matrices $M$ satisfying $M^{T}JM=J$ for the standard $J$. It proves basic properties including $\det M=1$, the inverse formula $M^{-1}=-J{M}^{T}J$, and the group structure. The Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$ is partially developed.

What is missing from Mathlib:

1. Williamson's theorem and symplectic eigenvalues: the diagonalisation of positive-definite matrices under symplectic conjugation, the uniqueness of the normal form, and the symplectic eigenvalue as the positive square root of an eigenvalue of $-JSJS$.
2. The Iwasawa decomposition $M=KAN$: the unique factorisation of a symplectic matrix into a rotation, a squeeze, and a shear.
3. The Maslov index and the topology of $\mathrm{Sp}(2n,\mathbb{R})$: specifically ${\pi }_1(\mathrm{Sp}(2n,\mathbb{R}))\cong \mathbb{Z}$ for $n\ge 1$.
4. The Lagrangian Grassmannian $\Lambda (n)\cong \mathrm{U}(n)/\mathrm{O}(n)$ as a homogeneous space and its cohomology $H^1(\Lambda (n),\mathbb{Z})\cong \mathbb{Z}$.
5. The metaplectic group $\mathrm{Mp}(2n,\mathbb{R})$ as the double cover of $\mathrm{Sp}(2n,\mathbb{R})$ and its representation on $L^2({\mathbb{R}}^n)$.
6. The classification of linear symplectic maps into elliptic, hyperbolic, and loxodromic types.

Items (1) and (2) are the most accessible formalisation targets. Williamson's theorem requires substantial development in symplectic linear algebra but uses only finite-dimensional tools. The metaplectic representation requires infinite-dimensional functional analysis. No Lean module ships with this unit.

Advanced results Master

The metaplectic representation

The symplectic group $\mathrm{Sp}(2n,\mathbb{R})$ does not have a faithful unitary representation on $L^2({\mathbb{R}}^n)$: it is a non-compact Lie group and the Stone-von Neumann theorem constrains its projective representations. However, its double cover -- the metaplectic group $\mathrm{Mp}(2n,\mathbb{R})$ -- admits a faithful unitary representation, the Segal-Shale-Weil representation (also called the oscillator representation or the metaplectic representation).

The construction proceeds as follows. For $X\in \mathfrak{sp}(2n,\mathbb{R})$, define the quadratic operator ${\hat{Q}}_X$ on Schwartz functions by replacing $(q,p)$ with the operators $(\hat{q},\hat{p})$ in the quadratic form $X$ encodes. The map $X\mapsto {\hat{Q}}_X$ is a Lie-algebra homomorphism from $\mathfrak{sp}(2n,\mathbb{R})$ into the skew-adjoint operators on $L^2({\mathbb{R}}^n)$ (modulo the factor $1/(2i\hslash )$). Exponentiating gives a projective representation of $\mathrm{Sp}(2n,\mathbb{R})$, which lifts to a genuine representation of the double cover $\mathrm{Mp}(2n,\mathbb{R})$.

Concretely, the generators act as:

• Rotation in the $(q_i,p_i)$ plane: $\hat{Q}=\frac{1}{2}({\hat{p}}_i^2+{\hat{q}}_i^2)$, the harmonic oscillator Hamiltonian. Its exponential is the harmonic oscillator propagator.
• Squeeze $q_i\to \lambda q_i$, $p_i\to p_i/\lambda$: $\hat{Q}=\frac{1}{2}({\hat{q}}_i{\hat{p}}_i+{\hat{p}}_i{\hat{q}}_i)$. Its exponential is the squeezing operator $\hat{S}(\lambda )$ of quantum optics.
• Shear $q_i\to q_i+\alpha p_i$: $\hat{Q}=\frac{\alpha }{2}{\hat{p}}_i^2$. Its exponential is a free-particle propagator over "time" $\alpha$.

The Fourier transform on ${\mathbb{R}}^n$ is (up to phases) the image of the rotation by $\pi /2$ in $\mathrm{Sp}(2n,\mathbb{R})$: the exchange $(q,p)\to (p,-q)$.

Quantum harmonic oscillator symmetries

The harmonic oscillator Hamiltonian $\hat{H}=\frac{1}{2}({\hat{p}}^2+{\omega }^2{\hat{q}}^2)$ has $\mathrm{U}(1)$ symmetry generated by $\hat{H}$ itself: the time-evolution $e^{-i\hat{H}t/\hslash }$ rotates the Wigner function in phase space. But the larger symmetry group is $\mathrm{SU}(1,1)\cong \mathrm{Sp}(2,\mathbb{R})/\{\pm I\}$, acting on the oscillator through the metaplectic representation. This includes:

• Time evolution (rotations in phase space),
• Squeezing (deforming the circular ground state into an ellipse),
• Squeezing + rotation (producing rotated squeezed states).

The full set of Gaussian states of the harmonic oscillator (ground state, coherent states, squeezed states, squeezed coherent states) is a single orbit of $\mathrm{Mp}(2,\mathbb{R})$ acting on the ground state. The stabiliser of the ground state is $\mathrm{U}(1)$ (time evolution), so the space of Gaussian states is the symmetric space $\mathrm{Sp}(2,\mathbb{R})/\mathrm{U}(1)\cong \mathrm{SU}(1,1)/\mathrm{U}(1)$, the hyperbolic plane. The geometry of Gaussian states is hyperbolic, with the symplectic form providing the metric.

For $n$ degrees of freedom, the symmetry group is $\mathrm{Mp}(2n,\mathbb{R})$, and the Gaussian states form an orbit of dimension $n(2n+1)-n^2=n(n+1)$, the dimension of the symmetric space $\mathrm{Sp}(2n,\mathbb{R})/\mathrm{U}(n)$.

The Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$ in detail

The Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$ has a natural decomposition into three subalgebras corresponding to the Iwasawa decomposition:

$$\mathfrak{sp}(2n,\mathbb{R})=\mathfrak{k}\oplus \mathfrak{a}\oplus \mathfrak{n}$$

where $\mathfrak{k}=\mathfrak{u}(n)$ (the skew-Hermitian matrices, dimension $n^2$), $\mathfrak{a}$ is the $n$-dimensional abelian subalgebra of diagonal squeezes, and $\mathfrak{n}$ is the $n(n-1)/2$-dimensional nilpotent subalgebra of upper-triangular shears. The dimensions add: $n^2+n+n(n-1)/2=n(2n+1)$.

The root system of $\mathfrak{sp}(2n,\mathbb{R})$ is of type $C_n$. The positive roots are $\{e_i-e_j,e_i+e_j,2e_i\}$ for $1\le i<j\le n$, giving $n^2$ positive roots. The Weyl group is the hyperoctahedral group of order $2^{n}n!$.

Connections to optics: ABCD matrices

In paraxial geometrical optics, a ray is characterised by its transverse position $q$ and its paraxial angle $p$ (the slope of the ray relative to the optical axis). An optical element (lens, mirror, free-space propagation, gradient-index medium) transforms $(q,p)$ by a $2\times 2$ matrix with determinant 1 -- an element of $\mathrm{Sp}(2,\mathbb{R})$.

The standard ABCD matrix is $M=\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)$ with $AD-BC=1$. This is exactly the symplectic condition for $n=1$. Common elements:

Element	ABCD matrix	Type
Free propagation, distance $d$	$\left(\begin{array}{cc}1& d\\ 0& 1\end{array}\right)$	Shear
Thin lens, focal length $f$	$\left(\begin{array}{cc}1& 0\\ -1/f& 1\end{array}\right)$	Shear
Magnifier, magnification $m$	$\left(\begin{array}{cc}m& 0\\ 0& 1/m\end{array}\right)$	Squeeze
Fourier-transform lens	$\left(\begin{array}{cc}0& f\\ -1/f& 0\end{array}\right)$	Rotation by $\pi /2$ + squeeze

The composition of optical elements corresponds to matrix multiplication in $\mathrm{Sp}(2,\mathbb{R})$. For $n$ transverse dimensions (e.g., fibre optics with cylindrically symmetric elements), the relevant group is $\mathrm{Sp}(4,\mathbb{R})$ or $\mathrm{Sp}(2,\mathbb{R})$ with appropriate symmetry constraints.

The symplectic structure of optics was first recognised by R. K. Luneburg (Mathematical Theory of Optics, 1964) and developed systematically by A. J. Dragt and others in the context of accelerator physics (Lie methods for particle beams). The connection to Hamiltonian mechanics is that the eikonal equation of geometrical optics is the high-frequency limit of the wave equation, just as classical mechanics is the $\hslash \to 0$ limit of quantum mechanics.

Synthesis. The symplectic group $\mathrm{Sp}(2n,\mathbb{R})$ is the algebraic skeleton underlying every linear canonical transformation. Its three generators -- rotations, squeezes, and shears -- are the building blocks from which every symplectic matrix is assembled (Iwasawa decomposition). Its eigenvalue constraints (reciprocal pairs) restrict the possible dynamics of linearised Hamiltonian systems. Its topology (${\pi }_1\cong \mathbb{Z}$) gives rise to the Maslov index and the metaplectic double cover. Williamson's theorem guarantees that any quadratic Hamiltonian can be symplectically diagonalised into harmonic oscillators. The metaplectic representation carries the classical symplectic symmetry into quantum mechanics, where the harmonic oscillator, squeezed states, and the Fourier transform are all manifestations of $\mathrm{Sp}(2n,\mathbb{R})$ acting on wave functions. In optics, the same group governs the propagation of Gaussian beams through optical systems. The central insight is that the symplectic group is the universal symmetry group of linear Hamiltonian systems, and its structure constrains everything from orbital stability to quantum state geometry to laser beam propagation.

Full proof set Master

Proposition 1 (Determinant equals $+1$). Every $M\in \mathrm{Sp}(2n,\mathbb{R})$ satisfies $\det M=+1$.

Proof. From $M^{T}JM=J$, take determinants: $\det (M^T)\det (J)\det (M)=\det (J)$. Since $\det (M^T)=\det (M)$ and $\det (J)=1$ (compute by row expansion or by noting $J$ is a permutation matrix with sign $(-1{)}^{n(n-1)/2}$ and $J^2=-I_n$ has $\det (J^2)=(-1{)}^n$, giving $(\det J{)}^2=(-1{)}^n$; but direct computation for the standard block form gives $\det J=1$), we get $(\det M{)}^2=1$, so $\det M=\pm 1$.

The identity matrix is symplectic with $\det I=1$. The map $M\mapsto \det M$ is continuous and takes only values $\pm 1$ on $\mathrm{Sp}(2n,\mathbb{R})$. By the polar decomposition, every symplectic matrix $M$ can be written as $M=PO$ where $P$ is positive-definite symplectic and $O\in \mathrm{Sp}(2n,\mathbb{R})\cap \mathrm{O}(2n)\cong \mathrm{U}(n)$. Since $\mathrm{U}(n)$ is connected and contains $I$, $\det O=|\det U{|}^2=1$ for $U\in \mathrm{U}(n)$. The positive-definite symplectic matrices form a convex cone (hence connected) containing $I$ with $\det I=1$. Therefore $\det M=1$ for all $M\in \mathrm{Sp}(2n,\mathbb{R})$. $\square$

Proposition 2 (Exponential of Lie-algebra element is symplectic). For any $X\in \mathfrak{sp}(2n,\mathbb{R})$ (i.e., $JX+X^{T}J=0$), $\exp (X)\in \mathrm{Sp}(2n,\mathbb{R})$.

Proof. Define $M(t)=\exp (tX)$. Compute:

$$\frac{d}{dt}\left[M(t{)}^{T}JM(t)\right]=M(t{)}^T{X}^{T}JM(t)+M(t{)}^{T}JXM(t)=M(t{)}^T\left(X^{T}J+JX\right)M(t)=0,$$

using the Lie-algebra condition $X^{T}J+JX=0$. So $M(t{)}^{T}JM(t)$ is constant in $t$. At $t=0$: $M(0)=I$, $I^{T}JI=J$. Hence $M(t{)}^{T}JM(t)=J$ for all $t$. Setting $t=1$: $\exp (X{)}^{T}J\exp (X)=J$. $\square$

Proposition 3 (Inverse formula). For $M\in \mathrm{Sp}(2n,\mathbb{R})$, $M^{-1}=-J{M}^{T}J$.

Proof. From $M^{T}JM=J$, multiply on the left by $J^{-1}$ and on the right by $M^{-1}$: $J^{-1}{M}^{T}JM{M}^{-1}=J^{-1}J{M}^{-1}$, which simplifies to $J^{-1}{M}^{T}J=M^{-1}$. Since $J^{-1}=-J$, this gives $M^{-1}=-J{M}^{T}J$. $\square$

Proposition 4 (Williamson's theorem for diagonal $S$). If $S=\mathrm{diag}(s_1,s_2)$ with $s_1,s_2>0$, then the symplectic eigenvalue of $S$ is $\lambda =\sqrt{s_1{s}_2}$, and $M^{T}SM=\lambda I_2$ for $M=\left(\begin{array}{cc}(s_2/s_1{)}^{1/4}& 0\\ 0& (s_1/s_2{)}^{1/4}\end{array}\right)R(\pi /4)$.

Proof. The matrix $JS=\left(\begin{array}{cc}0& s_2\\ -s_1& 0\end{array}\right)$ has eigenvalues $\pm i\sqrt{s_1{s}_2}$, so the symplectic eigenvalue is $\lambda =\sqrt{s_1{s}_2}$. The matrix $M$ is a squeeze followed by a $45^{\circ }$ rotation. Direct computation: $R(\pi /4{)}^{T}S(\alpha {)}^{T}SS(\alpha )R(\pi /4)$ with $S(\alpha )=\mathrm{diag}(\alpha ,1/\alpha )$ and $\alpha =(s_2/s_1{)}^{1/4}$ gives $\sqrt{s_1{s}_2}{I}_2$. Since both $S(\alpha )$ and $R(\pi /4)$ are symplectic, so is their product. $\square$

Connections Master

• Symplectic structure 09.04.05 introduces the symplectic form $\omega$ and Darboux's theorem; $\mathrm{Sp}(2n,\mathbb{R})$ is the group of linear transformations preserving the coordinate expression of $\omega$. The Darboux theorem says the nonlinear version of this preservation always holds locally.

• Canonical transformations 09.05.01 are the nonlinear generalisation of symplectic matrices. A linear canonical transformation is precisely an element of $\mathrm{Sp}(2n,\mathbb{R})$.

• Generating functions 09.05.03 provide a constructive tool for nonlinear canonical transformations. For linear canonical transformations, the generating function is quadratic: $F_2(q,P)=\frac{1}{2}{q}^{T}Aq+q^{T}BP+\frac{1}{2}{P}^{T}CP$, and the symplectic matrix is determined by the coefficients $A,B,C$.

• Liouville's theorem (phase-space volume preservation) follows from $\det M=1$ at the linear level. The fact that $\mathrm{Sp}(2n,\mathbb{R})\subset \mathrm{SL}(2n,\mathbb{R})$ (det = 1) is the linear shadow of the general fact that Hamiltonian flows preserve ${\omega }^n$.

• Normal modes 09.02.04 are found by symplectically diagonalising the quadratic Hamiltonian near an equilibrium. Williamson's theorem guarantees this is always possible.

• KAM theorem 09.08.01 requires the linearised Poincare map at an invariant torus to be symplectic. The eigenvalue constraints of $\mathrm{Sp}(2n,\mathbb{R})$ (reciprocal pairs on or off the unit circle) restrict the possible bifurcation scenarios.

Historical and philosophical context Master

The symplectic group was first studied systematically by Hermann Weyl in The Classical Groups: Their Invariants and Representations (Princeton, 1939). Weyl coined the term "symplectic" from the Greek $\sigma \upsilon \mu \pi \lambda ϵ\kappa \tau \iota \kappa \acute{o}\varsigma$ ("intertwined"), replacing the Latin root "complex" to avoid confusion with complex numbers. Weyl's treatment placed the symplectic group alongside the orthogonal and unitary groups as one of the three great families of classical Lie groups.

Carl Ludwig Siegel developed the symplectic theory in a different direction through his work on the symplectic geometry of the Siegel upper half-space ${\mathfrak{H}}_n=\{Z\in M_n(\mathbb{C}):Z=Z^T,\mathrm{Im}(Z)>0\}$, which is the quotient $\mathrm{Sp}(2n,\mathbb{R})/\mathrm{U}(n)$. Siegel's Symplectic Geometry (1943) and his later Lectures on the Geometry of Numbers (1989, written with K. Chandrasekharan) established the connection between symplectic groups and number theory (theta functions, modular forms). The Siegel upper half-space generalises the Poincare upper half-plane from $\mathrm{SL}(2,\mathbb{R})$ to $\mathrm{Sp}(2n,\mathbb{R})$ and is the natural domain for the Siegel theta function.

John Williamson (1936, 1937, 1939) proved the normal-form theorem for quadratic Hamiltonians under symplectic conjugation in a series of papers in the American Journal of Mathematics. Williamson's classification of normal forms into elliptic, hyperbolic, and loxodromic types remained relatively unknown until its rediscovery by the symplectic topology community in the 1980s. The theorem is now central to both classical mechanics (linearised stability analysis) and quantum optics (Gaussian states and their symplectic invariants).

The metaplectic representation was constructed independently by Irving Segal (1959), D. Shale (1962), and Andre Weil (1964), though its roots go back to van Hove (1951). Weil's treatment emphasised the connection to the Stone-von Neumann theorem and to theta functions. The double cover $\mathrm{Mp}(2n,\mathbb{R})\to \mathrm{Sp}(2n,\mathbb{R})$ is the simplest example of a metalinear group, and the non-existence of a genuine (as opposed to projective) representation of $\mathrm{Sp}(2n,\mathbb{R})$ on $L^2({\mathbb{R}}^n)$ is a consequence of the topology: ${\pi }_1(\mathrm{Sp}(2n,\mathbb{R}))\cong \mathbb{Z}$ prevents lifting to a linear representation without passing to the double cover.

The Maslov index was introduced by Victor Maslov in Theory of Perturbations and Asymptotic Methods (1965) and given its rigorous cohomological interpretation by V. I. Arnold in "On a characteristic class entering into conditions of quantisation" (1967). The Maslov index is the unique generator of $H^1(\Lambda (n),\mathbb{Z})\cong \mathbb{Z}$ and measures the topological obstruction to choosing a consistent phase for the WKB wave function as one traverses a path in the Lagrangian Grassmannian.

The ABCD matrix formalism in optics was systematised by R. K. Luneburg (Mathematical Theory of Optics, 1964) and later connected to the symplectic group by A. J. Dragt and E. Forest in the context of particle accelerator beam dynamics (1980s). The recognition that ray optics is a Hamiltonian system with symplectic symmetry goes back to William Rowan Hamilton himself, who developed his optical theory first (1827--1833) and then extended it to mechanics (1833--1834).

The philosophical significance of the symplectic group is that it is the linear version of a much larger nonlinear symmetry. Just as the rotation group $\mathrm{O}(n)$ is the linear symmetry of Euclidean geometry and the Lorentz group $\mathrm{O}(1,3)$ is the linear symmetry of spacetime, $\mathrm{Sp}(2n,\mathbb{R})$ is the linear symmetry of phase space. The nonlinear symplectomorphisms extend this symmetry to curved phase spaces, but the linear theory already captures the essential structure: the pairing of position and momentum, the preservation of oriented area, the constraints on eigenvalues, and the connection between classical symmetries and quantum operators through the metaplectic representation.

Bibliography Master

• Weyl, H., The Classical Groups: Their Invariants and Representations (Princeton University Press, 1939). The systematic treatment of the symplectic group alongside the orthogonal and unitary families. Coined the name "symplectic."

• Siegel, C. L., "Symplectic Geometry," Amer. J. Math. 65 (1943), 1--86. The Siegel upper half-space and its symplectic automorphisms.

• Williamson, J., "On the algebraic problem concerning the normal forms of linear dynamical systems," Amer. J. Math. 58 (1936), 141--163. The normal-form theorem for quadratic Hamiltonians.

• Arnold, V. I., "On a characteristic class entering into conditions of quantisation," Funkts. Anal. Prilozh. 1 (1967), 1--14. The Maslov index as a cohomology class.

• Maslov, V. P., Theorie des perturbations et methodes asymptotiques (Dunod, 1972). The Maslov index and singular generating functions.

• Segal, I. E., "Foundations of the theory of dynamical systems of infinitely many degrees of freedom I," Mat.-Fys. Medd. Danske Vid. Selsk. 31 (1959), no. 12. The metaplectic representation.

• Shale, D., "Linear symmetries of free Boson fields," Trans. Amer. Math. Soc. 103 (1962), 149--167. Independent construction of the metaplectic representation.

• Weil, A., "Sur certains groupes d'operateurs unitaires," Acta Math. 111 (1964), 143--211. The metaplectic representation and its connection to theta functions.

• Arnold, V. I., Mathematical Methods of Classical Mechanics, 2nd ed. (Springer GTM 60, 1989), §38.

• McDuff, D. & Salamon, D., Introduction to Symplectic Topology, 3rd ed. (Oxford University Press, 2017), Ch. 1.

• Goldstein, H., Poole, C. P. & Safko, J. L., Classical Mechanics, 3rd ed. (Pearson, 2002), Ch. 9.

• Landau, L. D. & Lifshitz, E. M., Mechanics, 3rd ed. (Pergamon, 1976), §45.

• Luneburg, R. K., Mathematical Theory of Optics (University of California Press, 1964). ABCD matrices as symplectic transformations.

• Dragt, A. J. & Forest, E., "Lie algebraic theory of charged-particle optics and electron microscopes," Adv. Electron. Electron Phys. 67 (1986), 65--120. Symplectic methods in accelerator physics.