Symplectic Condition

By this point in the chapter you own two ways of certifying a canonical transformation: exhibit a generating function, or grind through Poisson brackets pair by pair. Both work; neither scales gracefully. The symplectic condition compresses the entire question into one matrix identity. Stack the phase-space variables into a column $\eta=(q_1,\dots,q_n,p_1,\dots,p_n)^{\mathsf T}$ , write the transformation's Jacobian $M$ , and ask a single question: does $M^{\mathsf T}JM=J$ hold, where $J$ is the fixed antisymmetric matrix that encodes "who is conjugate to whom"? If yes, the map is canonical. Every consequence you have met — and several you have not — falls out of this identity by matrix algebra: $\det M=1$ in one line (Liouville's theorem), preservation of all fundamental Poisson brackets (they are literally the entries of $MJM^{\mathsf T}$ ), and the closure of canonical maps under composition and inversion, which promotes them to a group. The condition also changes how you should see phase space. Orthogonal matrices satisfy $R^{\mathsf T}IR=I$ and preserve lengths; symplectic matrices satisfy $M^{\mathsf T}JM=J$ and preserve oriented conjugate-pair areas. Phase space carries no meaningful notion of distance — stretch it, shear it freely — but the signed area a trajectory's neighbors enclose in each $(q_i,p_i)$ plane is sacred. That antisymmetric geometry, not any metric, is what Hamiltonian mechanics actually runs on. This page derives the condition from Hamilton's equations in three moves, proves its equivalence to the bracket test, checks it on numbers, and extracts its sharpest physical payoff: the eigenvalues of any Hamiltonian stability matrix come locked in reciprocal pairs.

Key takeaways

A phase-space transformation $(q,p)\to(Q,P)$ is canonical iff its Jacobian satisfies the symplectic condition $M^{\mathsf T}JM=J$ (equivalently $MJM^{\mathsf T}=J$ ), with $J$ the constant antisymmetric symplectic form.
The condition is Hamilton's-equations-preservation written as matrix algebra: $\dot\eta=J\,\partial_\eta H$ transforms to $\dot\zeta=MJM^{\mathsf T}\partial_\zeta H$ , so form-invariance demands $MJM^{\mathsf T}=J$ .
It is identical in content to preservation of the fundamental Poisson brackets: $\{\zeta_i,\zeta_j\}_\eta=(MJM^{\mathsf T})_{ij}$ .
Taking determinants gives $(\det M)^2=1$ , and a finer Pfaffian argument forces $\det M=+1$ : phase-space volume is exactly preserved — Liouville's theorem in one line.
Symplectic Jacobians close under products and inverses: they form the symplectic group $Sp(2n,\mathbb{R})$ , and Hamiltonian flow itself is a one-parameter family inside it.
Eigenvalues of symplectic matrices come in reciprocal pairs $(\lambda,1/\lambda)$ — the rigidity behind Hamiltonian stability theory and parametric resonance.
Key idea: canonicity is geometry — a transformation is canonical iff its Jacobian preserves the antisymmetric form $J$ , all of it packaged in the single identity $M^{\mathsf T}JM=J$ .

The core insight

Canonicity is geometry: a transformation is canonical iff its Jacobian preserves the symplectic 2-form $J$ , encoded in the single matrix identity $M^{\mathsf T}JM=J$ .

Where this lives today

The symplectic condition is a production requirement in accelerator physics. A particle in the LHC or a synchrotron light source completes upward of $10^{9}$ revolutions; its lattice is modeled as a chain of transfer matrices, and if numerical truncation makes any of them even slightly non-symplectic, simulated beams spiral into artificial growth or decay that no real magnet produces. Design codes therefore symplectify — they project computed maps back onto $Sp(2n)$ before tracking, and the stability analysis of the ring is exactly worked example 2's trace criterion applied to the one-turn matrix. The same identity powers the long-time solar-system integrations behind planetary chronology, and modern machine learning has joined in: symplectic neural networks hard-wire $M^{\mathsf T}JM=J$ into their layers so that learned surrogate dynamics cannot leak phase-space volume. Meanwhile Gromov's 1985 non-squeezing theorem sharpened the geometry itself: a canonical map cannot squeeze a ball through a hole narrower than its conjugate-pair cross-section, no matter how volume-preserving — a purely classical "uncertainty principle" that quantum mechanics, via $[\hat q,\hat p]=i\hbar$ , was foreshadowing all along.

The physical picture

Forget lengths and angles: phase space does not have them. What it has is oriented area, counted conjugate pair by conjugate pair. Take a small parallelogram of nearby states spanned by displacement vectors $\delta\eta_1$ , $\delta\eta_2$ ; the quantity $\delta\eta_1^{\mathsf T}J\,\delta\eta_2$ sums the signed areas of the parallelogram's shadows on the $(q_1,p_1)$ plane, the $(q_2,p_2)$ plane, and so on. That number — not any distance — is what a canonical transformation must leave untouched. The symplectic condition is exactly this statement made algebraic: $\,(M\delta\eta_1)^{\mathsf T}J(M\delta\eta_2)=\delta\eta_1^{\mathsf T}(M^{\mathsf T}JM)\delta\eta_2$ equals the old area for all parallelograms iff $M^{\mathsf T}JM=J$ .

Figure

Two geometries — rotations preserve lengths against the metric I, symplectic maps preserve oriented conjugate-pair areas against J

The switch from a symmetric invariant ( $I$ ) to an antisymmetric one ( $J$ ) changes the character of the group completely. Rotations form a compact group — you can only turn so far. Symplectic maps form a non-compact group: arbitrarily violent shears and stretches are allowed, provided conjugate areas balance. That is why chaotic Hamiltonian dynamics can stretch a blob of states into an absurd filament — while Liouville's theorem holds exactly, because filamentation is area-rebalancing, not area-creation. The payoff picture is about stability. Linearize any Hamiltonian flow around a periodic orbit; the once-around Jacobian (monodromy matrix) is symplectic, and symplectic eigenvalues come in reciprocal pairs $(\lambda,1/\lambda)$ . Consequence: an eigenvalue cannot drift off the unit circle alone — the moment one leaves (instability), its reciprocal partner leaves in the opposite direction. Hamiltonian systems cannot be "slightly damped" into stability or out of it; they sit forever on the knife edge that makes parametric resonance, KAM theory, and accelerator design the subtle subjects they are.

Figure

Symplectic eigenvalue rigidity — eigenvalues live in reciprocal pairs, so they can leave the unit circle only by colliding and splitting

From Hamilton's equations to one matrix identity

First, compress the canonical equations. With $\eta=(q,p)^{\mathsf T}$ and the block matrix

J=\begin{pmatrix}0_n & I_n\\ -I_n & 0_n\end{pmatrix},\qquad J^{\mathsf T}=-J,\quad J^{2}=-I_{2n},\quad J^{-1}=-J \qquad (1)

where: $I_n$ , $0_n$ — identity and zero $n\times n$ blocks. $J$ is the algebra of "conjugateness" itself: applied to a gradient, it swaps coordinate and momentum slots with one sign flip — exactly the crossover pattern of Hamilton's equations, which become a single line:

\dot\eta=J\,\frac{\partial H}{\partial\eta} \qquad (2)

where: $\partial H/\partial\eta$ — the $2n$ -gradient. Unpacking the blocks returns $\dot q=\partial H/\partial p$ , $\dot p=-\partial H/\partial q$ ; see Hamilton's Canonical Equations. Everything the symplectic condition says flows from asking: for which maps $\zeta(\eta)$ does (2) survive with $\eta\to\zeta$ ?

M_{ij}=\frac{\partial\zeta_i}{\partial\eta_j},\qquad M=\begin{pmatrix}\partial Q/\partial q & \partial Q/\partial p\\[2pt] \partial P/\partial q & \partial P/\partial p\end{pmatrix} \qquad (3)

where: $M$ — the Jacobian, evaluated pointwise (for nonlinear maps it varies over phase space; the condition must hold at every point). The answer (Derivation A):

M J M^{\mathsf T}=J \qquad\Longleftrightarrow\qquad M^{\mathsf T}JM=J \qquad (4)

where: the two forms are equivalent because $J$ is invertible and symplectic matrices are closed under transpose. This is the symplectic condition. One matrix identity, $n(2n+1)$ independent scalar constraints — exactly enough to cut the $2n\times2n$ Jacobians down to "half the freedom," matching the one-scalar-function content of a generating function.

Derivation AThe symplectic condition from form-invariance of Hamilton's equations

We are deriving the master criterion: a (time-independent) transformation $\zeta=\zeta(\eta)$ preserves the canonical form of Hamilton's equations for every Hamiltonian iff its Jacobian satisfies $MJM^{\mathsf T}=J$ . The strategy is three applications of the chain rule — transform the velocities, substitute the old equations, transform the gradient — and then demand that what emerges is again equation (2) in the new variables. The subtle step is the gradient: it transforms with $M^{\mathsf T}$ , not $M$ , and that transpose is what makes the sandwich $MJM^{\mathsf T}$ appear.

Step 1. Transform the velocities. By the chain rule, $\dot\zeta_i=\sum_j(\partial\zeta_i/\partial\eta_j)\dot\eta_j$ , i.e.

\dot\zeta=M\,\dot\eta

Step 2. Substitute the old equations of motion, $\dot\eta=J\,\partial H/\partial\eta$ :

\dot\zeta=MJ\,\frac{\partial H}{\partial\eta}

Step 3. Transform the gradient. Viewing $H(\eta)=H(\eta(\zeta))$ as a function of the new variables, the chain rule gives $\partial H/\partial\eta_j=\sum_i(\partial\zeta_i/\partial\eta_j)\,\partial H/\partial\zeta_i$ , i.e.

\frac{\partial H}{\partial\eta}=M^{\mathsf T}\frac{\partial H}{\partial\zeta}

Why the transpose: velocities are pushed forward along the map (columns of $M$ ), gradients are pulled back (rows of $M$ ) — the standard contravariant/covariant split, the same one that made momenta transform inversely to coordinates in the point-transformation lift.

Step 4. Combine:

\dot\zeta=MJM^{\mathsf T}\,\frac{\partial H}{\partial\zeta}

This has Hamiltonian form in the new chart — for arbitrary $H$ — precisely when the sandwich equals $J$ : $\ MJM^{\mathsf T}=J$ .

Step 5. Convert to the transpose form. From $MJM^{\mathsf T}=J$ , multiply by $M^{-1}$ on the left and $(M^{\mathsf T})^{-1}$ on the right to get $J=M^{-1}J(M^{\mathsf T})^{-1}$ , i.e. $M^{-1}$ is symplectic; then writing the condition for $M^{-1}$ and inverting again yields $M^{\mathsf T}JM=J$ . The two forms carry identical content, and give the closed inversion formula $M^{-1}=-JM^{\mathsf T}J=J^{-1}M^{\mathsf T}J$ — no numerical matrix inversion ever needed for a symplectic matrix. Verified: reproduces Goldstein §9.4 (eqs. 9.55) with the restricted-transformation argument; block-expanding (4) for $n=1$ returns the single condition $\det M=1$ , checked against the known area-preservation criterion.

Two heavyweight corollaries drop out by pure matrix algebra. First, take determinants of (4): $\det M^{\mathsf T}\,\det J\,\det M=\det J$ , so

(\det M)^{2}=1\qquad\text{and in fact}\qquad \det M=+1 \qquad (5)

where: the exclusion of $-1$ is genuine mathematics, not convention — the Pfaffian identity $\mathrm{Pf}(M^{\mathsf T}JM)=\det(M)\,\mathrm{Pf}(J)$ pins the sign. Physically: canonical maps preserve phase-space volume exactly, and orientation. Since Hamiltonian evolution is itself canonical (Derivation C), the volume of any co-moving region is frozen — Liouville's theorem in one line of algebra; the full statistical-mechanics story lives in the Liouville article. Second, the condition is the Poisson-bracket test, not merely equivalent-in-spirit:

\{\zeta_i,\zeta_j\}_{\eta}=\left(MJM^{\mathsf T}\right)_{ij} \qquad (6)

where: $\{\,,\}_\eta$ — brackets computed in the old chart. The fundamental brackets of the new variables are the entries of the sandwich matrix. Demanding $\{Q_i,Q_j\}=\{P_i,P_j\}=0$ , $\{Q_i,P_j\}=\delta_{ij}$ is demanding $MJM^{\mathsf T}=J$ , entry by entry.

Go deeper · Derivation B · 1 minThe symplectic condition and the fundamental brackets are the same statement

Derivation BThe symplectic condition and the fundamental brackets are the same statement

We are proving that the matrix criterion and the Poisson-bracket test coincide exactly — one computation, two languages. The strategy: write the bracket of two phase-space functions in symplectic notation, specialize to the new coordinates themselves, and recognize the Jacobian sandwich. This derivation is the reason you never need to run both tests.

Step 1. The Poisson bracket in symplectic notation. For any $F(\eta)$ , $G(\eta)$ ,

\{F,G\}_{\eta}=\left(\frac{\partial F}{\partial\eta}\right)^{\!\mathsf T}\!J\,\frac{\partial G}{\partial\eta}

Block-expanding reproduces the familiar $\sum_i(\partial_qF\,\partial_pG-\partial_pF\,\partial_qG)$ — see Poisson Brackets: Definition.

Step 2. Specialize to the new coordinates: $F=\zeta_i$ , $G=\zeta_j$ . Their gradients are rows of the Jacobian, $\partial\zeta_i/\partial\eta_k=M_{ik}$ :

\{\zeta_i,\zeta_j\}_{\eta}=\sum_{k,l}M_{ik}\,J_{kl}\,M_{jl}=\left(MJM^{\mathsf T}\right)_{ij}

Step 3. Impose the fundamental-bracket requirement. The new chart is canonical iff $\{Q_i,Q_j\}=0$ , $\{P_i,P_j\}=0$ , $\{Q_i,P_j\}=\delta_{ij}$ — which, assembled into a $2n\times2n$ antisymmetric array, is precisely the matrix $J$ . Hence canonicity $\iff MJM^{\mathsf T}=J$ . Why this settles the redundancy question: of the $4n^2$ bracket conditions, antisymmetry leaves only $n(2n+1)$ independent ones — exactly the number of scalar equations in the matrix identity. Neither test carries information the other lacks. Verified: reproduces Goldstein §9.4 eqs. (9.55)–(9.62); for the exchange map $Q=p$ , $P=-q$ the sandwich gives $MJM^{\mathsf T}=JJ(-J)\cdot$ (sign bookkeeping) $=J$ , consistent with its bracket $\{Q,P\}=1$ computed directly in the type-3/4 article.

Go deeper · Derivation C · 1 minInfinitesimal maps, the symplectic algebra, and why Hamiltonian flow is canonical

Derivation CInfinitesimal maps, the symplectic algebra, and why Hamiltonian flow is canonical

We are answering the dynamical question: the flow generated by any Hamiltonian — the map "state now $\mapsto$ state $\mathrm{d}t$ later," composed over finite times — satisfies the symplectic condition. The strategy: expand a near-identity map's Jacobian to first order, discover that the condition constrains the generator's Hessian to enter as $J\times(\text{symmetric})$ , and recognize Hamilton's equations as exactly such a flow. This also exposes the Lie algebra $\mathfrak{sp}(2n)$ : the "velocity space" of the symplectic group.

Step 1. A near-identity transformation with generator $G(\eta)$ (the type-2 construction of Type-2 Generating Function F2(q,P,t), written symplectically):

\zeta=\eta+\epsilon\,J\,\frac{\partial G}{\partial\eta}

Step 2. Its Jacobian, to first order:

M=I+\epsilon\,J S,\qquad S_{ij}=\frac{\partial^{2}G}{\partial\eta_i\,\partial\eta_j}=S_{ji}

Why $S$ is symmetric: it is a Hessian — mixed partials of one scalar commute. This symmetry is the entire engine of the proof.

Step 3. Check the condition to first order, using $J^{\mathsf T}=-J$ so that $M^{\mathsf T}=I-\epsilon SJ$ :

M^{\mathsf T}JM=(I-\epsilon SJ)\,J\,(I+\epsilon JS)=J+\epsilon\left(J^{2}S-SJ^{2}\right)+O(\epsilon^{2})

Step 4. Apply $J^{2}=-I$ : the first-order piece is $-S+S=0$ , leaving $M^{\mathsf T}JM=J+O(\epsilon^{2})$ . Matrices of the form $JS$ with symmetric $S$ constitute the symplectic Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$ ; exponentiating them produces finite symplectic matrices, and the $O(\epsilon^2)$ error vanishes in the composition limit exactly as in the exponential map.

Step 5. Set $G=H$ and $\epsilon=\mathrm{d}t$ : the infinitesimal map is $\dot\eta=J\,\partial_\eta H$ — Hamiltonian evolution. Composing $t/\mathrm{d}t$ such steps, the finite-time flow map is a product of symplectic matrices, hence symplectic (group closure). Time evolution is a one-parameter subgroup of $Sp(2n,\mathbb{R})$ — where the story of Symplectic Group and Phase-Space Geometry picks up. Verified: consistent with Arnold Ch. 8 and the standard characterization $\mathfrak{sp}(2n)=\{JS:S=S^{\mathsf T}\}$ ; dimension check — symmetric $S$ has $n(2n+1)$ free entries, matching the group dimension.

Limiting cases

Case	Result	Meaning
One degree of freedom	condition $\iff\det M=\partial_qQ\,\partial_pP-\partial_pQ\,\partial_qP=1$	for $n=1$ symplectic = area-preserving; a single scalar check
$n\geq2$	strictly stronger than $\det M=1$	volume preservation is necessary, not sufficient — each conjugate-pair area is separately constrained
Point transformation	$Q=Q(q)$ , $P=(\partial q/\partial Q)^{\mathsf T}p$	coordinate changes lift canonically iff momenta transform contragrediently; Jacobian then symplectic automatically
Linear Hamiltonian flow	$M(t)=\exp(JAt)$ , $A=A^{\mathsf T}$	quadratic Hamiltonians generate one-parameter symplectic families — the oscillator's rotation matrix among them
Scaling one variable	$Q=\lambda q$ , $P=p$ : $M^{\mathsf T}JM=\lambda J\neq J$	fails; only the paired scaling $Q=\lambda q$ , $P=p/\lambda$ is canonical
Reflection $Q=-q$ , $P=p$	$\det M=-1$	anticanonical: orientation reversed, brackets flip sign — excluded by the Pfaffian argument

Common trapdet M equals 1 certifies canonicity only in one dimension

The seductive shortcut — "Jacobian determinant is one, so the map is canonical" — is valid only for a single degree of freedom, where the symplectic condition happens to collapse to that one scalar. For $n\geq2$ the condition imposes $n(2n+1)$ constraints and $\det M=1$ is just one consequence of them: a map can preserve total phase-space volume while shuffling area between the $(q_1,p_1)$ and $(q_2,p_2)$ planes, and such a map is NOT canonical — it will wreck Hamilton's equations even though Liouville-style volume bookkeeping looks fine. (Example: $Q_1=2q_1$ , $P_1=p_1$ , $Q_2=q_2/2$ , $P_2=p_2$ has $\det M=1$ and $\{Q_1,P_1\}=2$ .) Always check the full sandwich $M^{\mathsf T}JM$ , or equivalently all fundamental brackets.

Worked examples

1 · One map passes, one map fails (confidence builder)

Test two linear maps at one degree of freedom. (a) $Q=2q+0.5p$ , $P=0.5p$ (positions in m, momenta in kg m/s). (b) $Q=2q$ , $P=p$ . Check the symplectic condition numerically for each.

Step 1. Map (a): the Jacobian is constant, $M=\begin{pmatrix}2&0.5\\0&0.5\end{pmatrix}$ . Compute the sandwich in two steps. First $JM=\begin{pmatrix}0&1\\-1&0\end{pmatrix}\begin{pmatrix}2&0.5\\0&0.5\end{pmatrix}=\begin{pmatrix}0&0.5\\-2&-0.5\end{pmatrix}$ .

Step 2. Then $M^{\mathsf T}JM=\begin{pmatrix}2&0\\0.5&0.5\end{pmatrix}\begin{pmatrix}0&0.5\\-2&-0.5\end{pmatrix}=\begin{pmatrix}0&1\\-1&0\end{pmatrix}=J$ ✓. (Entry check: top-right $=2\times0.5+0\times(-0.5)=1$ ; bottom-left $=0.5\times0-0.5\times2=-1$ .) Canonical — a stretch of $q$ by 2 paid for by shrinking the momentum direction, plus a harmless shear.

Step 3. Map (b): $M=\begin{pmatrix}2&0\\0&1\end{pmatrix}$ , so $M^{\mathsf T}JM=\begin{pmatrix}0&2\\-2&0\end{pmatrix}=2J\neq J$ . Not canonical: the bracket comes out $\{Q,P\}=2$ . Hamilton's equations in $(Q,P)$ would run at double speed with the substituted Hamiltonian — the classic silent factor-of-two bug in oscillator homework.

Step 4. Repair map (b) by paying the area debt: $Q=2q$ , $P=p/2$ gives $M=\mathrm{diag}(2,\tfrac12)$ , $\det M=1$ , and $M^{\mathsf T}JM=J$ ✓. At the phase point $q=0.30\ \mathrm{m}$ , $p=1.6\ \mathrm{kg\,m/s}$ : $(Q,P)=(0.60\ \mathrm{m},\,0.80\ \mathrm{kg\,m/s})$ , and the conjugate-pair area of any small patch around it is unchanged.

2 · The oscillator's own flow map is symplectic — and its eigenvalues tell the stability story (exam level)

The harmonic oscillator with $m=2.0\ \mathrm{kg}$ , $\omega=5.0\ \mathrm{rad/s}$ has the exact time- $t$ flow map $q(t)=q_0\cos\omega t+\dfrac{p_0}{m\omega}\sin\omega t$ , $\ p(t)=-m\omega q_0\sin\omega t+p_0\cos\omega t$ . At $\omega t=\pi/3$ , verify the symplectic condition numerically and analyze the eigenvalues of $M$ .

Step 1. Build the Jacobian at $\omega t=\pi/3$ ( $\cos=0.5$ , $\sin=0.86603$ , $m\omega=10.0\ \mathrm{kg/s}$ ):

M=\begin{pmatrix}\cos\omega t & \frac{\sin\omega t}{m\omega}\\[2pt] -m\omega\sin\omega t & \cos\omega t\end{pmatrix}=\begin{pmatrix}0.5 & 0.086603\\ -8.6603 & 0.5\end{pmatrix}

Step 2. Determinant first: $\det M=(0.5)(0.5)-(0.086603)(-8.6603)=0.25+0.75=1.000$ ✓. For $n=1$ this already settles canonicity, but run the sandwich to see the machinery: $JM=\begin{pmatrix}-8.6603&0.5\\-0.5&-0.086603\end{pmatrix}$ , then

M^{\mathsf T}JM=\begin{pmatrix}0.5&-8.6603\\0.086603&0.5\end{pmatrix}\begin{pmatrix}-8.6603&0.5\\-0.5&-0.086603\end{pmatrix}=\begin{pmatrix}0&1\\-1&0\end{pmatrix}=J\ \checkmark

(Top-left: $0.5\times(-8.6603)+(-8.6603)\times(-0.5)=0$ ; top-right: $0.5\times0.5+(-8.6603)(-0.086603)=0.25+0.75=1$ .) The flow of the dynamics itself passes the test at this — and every — instant.

Step 3. Eigenvalues. The characteristic equation is $\lambda^{2}-(\mathrm{Tr}\,M)\lambda+\det M=0$ with $\mathrm{Tr}\,M=1.0$ and $\det M=1$ :

\lambda=\frac{1.0\pm\sqrt{1.0-4.0}}{2}=0.5\pm0.86603\,i=e^{\pm i\pi/3}

A reciprocal pair on the unit circle: $\lambda\cdot\lambda^{*}=|\lambda|^{2}=1$ and $\lambda_2=1/\lambda_1$ ✓. The motion is stable — phase points circulate without growth, at angular rate $\omega$ exactly.

Step 4. Read the stability criterion off the trace. For any $2\times2$ symplectic $M$ , $\det M=1$ forces $\lambda=\tfrac12(\mathrm{Tr}M\pm\sqrt{\mathrm{Tr}^2M-4})$ : eigenvalues ride the unit circle iff $|\mathrm{Tr}\,M|<2$ . Here $|\mathrm{Tr}\,M|=1<2$ ✓ stable; had the "oscillator" been inverted ( $\omega^2<0$ , $\cosh/\sinh$ flow), the trace would exceed 2 and the pair would split to $(\lambda,1/\lambda)$ on the real axis — exponential growth paid by exponential decay, orientation and area still perfectly preserved. This trace test is, verbatim, the stability criterion used for particle orbits in storage rings and for Mathieu-type parametric resonance.

Origins · 1 minHermann Weyl and how this was discovered

Hermann Weyl

1885–1955 · Göttingen, Zürich, Princeton

The word symplectic is Weyl's. Needing a name for the group preserving an antisymmetric bilinear form, and finding "complex group" (his earlier coinage) hopelessly ambiguous, he built a Greek calque of the Latin complexus — braided together — and introduced "symplectic group" in The Classical Groups (1939), the book in which he worked out the invariant theory of $Sp(2n)$ alongside the orthogonal and linear families. It was a characteristic Weyl move: he spent his career finding the load-bearing structures beneath physics' formalisms. His Raum, Zeit, Materie (1918) put general relativity on a geometric footing and introduced gauge invariance — the idea, initially rejected by Einstein, that became the organizing principle of particle physics; his group-theoretic quantum mechanics (1927–28) recast canonical commutation relations as a representation problem, the quantum image of the symplectic structure on this page (today's "Weyl quantization" and the metaplectic representation descend from it). Colleague of Hilbert and Einstein at Göttingen and Zürich, founding member of the Institute for Advanced Study after fleeing Germany in 1933, Weyl held that his work always tried "to unite the true with the beautiful" — and when forced to choose, he chose the beautiful. In the symplectic condition, mechanics got both.

History

1809

— Lagrange and Poisson develop the bracket calculus while analyzing planetary perturbations — symplectic structure in embryo

1834

— Hamilton compresses dynamics into the canonical equations, the form the condition protects

1890

— Poincaré identifies the integral invariants of Hamiltonian flow: conjugate-pair areas conserved

1939

— Weyl coins "symplectic" and develops the group's invariant theory in The Classical Groups

1968

— Arnold's Mathematical Methods geometrizes mechanics: phase space as symplectic manifold

1985

— Gromov's non-squeezing theorem reveals symplectic rigidity far beyond volume preservation

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §9.4 (The Symplectic Approach to Canonical Transformations) · Landau & Lifshitz, Mechanics, 3rd ed., §45 · Morin, Introduction to Classical Mechanics, Ch. 15 (Hamiltonian formalism) · Taylor, Classical Mechanics, Ch. 13 · Arnold, Mathematical Methods of Classical Mechanics, Ch. 8

Provenance. Equations derived from Hamilton's equations and the standard symplectic form. A verifier should recompute: (1) the chain-rule passage $\dot\zeta=MJM^{\mathsf T}\partial H/\partial\zeta$ leading to $M^{\mathsf T}JM=J$ (Goldstein §9.4); (2) $(\det M)^2=1$ from the determinant of the condition; (3) the equivalence with fundamental Poisson brackets via $\{\zeta_i,\zeta_j\}=(MJM^{\mathsf T})_{ij}$ ; (4) the harmonic check $M=J$ in the SHO example; (5) the group axioms for $Sp(2n,\mathbb R)$ . Cited: Goldstein Ch. 9, Landau-Lifshitz §45, Arnold Ch. 8. The group dimension $n(2n+1)$ is the standard value for $Sp(2n,\mathbb R)$ . The numeric examples (linear pass/fail pair; oscillator flow at $\omega t=\pi/3$ with $m=2.0$ , $\omega=5.0$ ) were computed for this article; each matrix product is checkable by hand.

Builds on: Definition of Canonical Transformations · Generating Functions and the Four Types · Fundamental Poisson Brackets · Poisson Brackets: Definition Leads to: Symplectic Group and Phase-Space Geometry · Poisson Brackets as a Test for Canonicity · Poisson-Bracket Formulation of Dynamics · Algebra of Poisson Brackets

Are the two forms M-transpose-J-M and M-J-M-transpose really interchangeable? Some books use one, some the other.

Yes — for symplectic matrices they are equivalent, and the proof is two lines: from $MJM^{\mathsf T}=J$ you can solve $M^{-1}=-JM^{\mathsf T}J$ , substitute back, and obtain $M^{\mathsf T}JM=J$ ; the reverse runs identically. The underlying reason is that the inverse and the transpose of a symplectic matrix are themselves symplectic, so "M preserves J" and "M-transpose preserves J" are the same club membership. What you may not do is mix them mid-derivation without noticing which chain-rule convention (velocities forward vs gradients back) your Jacobian indices follow.

Where exactly does det M = −1 get excluded? Squaring gives both signs.

By the Pfaffian — a polynomial $\mathrm{Pf}(A)$ defined for antisymmetric matrices with $\mathrm{Pf}(A)^2=\det A$ and the transformation law $\mathrm{Pf}(M^{\mathsf T}AM)=\det(M)\,\mathrm{Pf}(A)$ . Apply it to the condition: $\mathrm{Pf}(J)=\det(M)\,\mathrm{Pf}(J)$ , and since $\mathrm{Pf}(J)=1\neq0$ , $\det M=+1$ exactly. Geometrically: a $\det=-1$ map would reverse phase-space orientation (like $q\to-q$ alone), flipping the sign of every bracket — the "anticanonical" maps discussed in the type-3/4 article. They preserve volume but not the symplectic form.

Does passing the symplectic condition depend on which Hamiltonian I plan to use?

No — and that is the condition's quiet strength. $M^{\mathsf T}JM=J$ is a property of the map alone; Derivation A demanded form-invariance for an arbitrary $H$ , and the Hamiltonian dropped out of the requirement entirely. One certified transformation works for every system: the polar-coordinates lift is canonical whether your potential is Kepler's or a galaxy's. (Contrast with weaker notions like "preserves the equations of this particular H," which some older texts call canonoid — such maps exist but transfer to no other Hamiltonian.)

For time-dependent transformations, what happens to the condition?

It survives untouched, applied to the spatial Jacobian: at each frozen instant $t$ , the matrix $M_{ij}=\partial\zeta_i/\partial\eta_j$ (partials at fixed $t$ ) must satisfy $M^{\mathsf T}JM=J$ . What time dependence changes is the energetics: the new Hamiltonian picks up the generator's clock term, $K=H+\partial F/\partial t$ , exactly as for every generating-function type. Geometry and bookkeeping separate cleanly — the map's symplectic character at each instant, the Hamiltonian's correction across instants.

Why do symplectic eigenvalues pair as λ and 1/λ, and why should I care?

From the condition: $M^{\mathsf T}=JM^{-1}J^{-1}$ , so $M^{\mathsf T}$ — which always shares eigenvalues with $M$ — is also similar to $M^{-1}$ , whose eigenvalues are the reciprocals. Hence the spectrum is closed under $\lambda\to1/\lambda$ (and under conjugation, being real). The physics: linearize around any orbit and the stability matrix is symplectic, so growth rates come in balancing pairs — a Hamiltonian system cannot have all its eigenvalues damped inside the unit circle. Attractors are impossible without dissipation; stability lives only on the knife edge of the unit circle. Every storage-ring design and every KAM argument leans on this rigidity.

Is being symplectic the same as being area-preserving?

In one degree of freedom, yes: the condition collapses to $\det M=1$ . In $n\geq2$ , symplectic is strictly stronger than volume-preserving — see the warning box — and since 1985 we know it is enormously stronger: Gromov's non-squeezing theorem shows a symplectic map cannot push a ball of radius $R$ into a cylinder of smaller conjugate cross-section $\pi r^2<\pi R^2$ , though a volume-preserving map does it trivially by squeezing one pair and stretching another. Symplectic geometry has genuine rigidity, halfway between floppy volume geometry and rigid metric geometry.

How do I actually run this test on a nonlinear transformation?

Pointwise. Compute the Jacobian symbolically as a function of $(q,p)$ , form $M^{\mathsf T}JM$ , and simplify — the result must equal $J$ identically in the variables, not just at a special point. In practice: for $n=1$ just verify $\{Q,P\}=1$ symbolically (one bracket); for larger $n$ the matrix sandwich organizes the $n(2n+1)$ bracket checks so cancellations are visible. If you generated the map from any $F_1$ – $F_4$ , skip the test — Derivation B of the type-1 article and its siblings guarantee it identically.

Which is more fundamental — the symplectic condition, Poisson brackets, or generating functions?

They are one structure in three costumes: the condition is the local, algebraic face (a constraint on Jacobians), brackets are the observable face (what the constraint means for measurable quantities), and generating functions are the constructive face (how to manufacture maps satisfying it). Modern geometry takes the two-form $\omega=\mathrm{d}q_i\wedge\mathrm{d}p_i$ (whose matrix is $J$ ) as the primitive object: a symplectic manifold is a space equipped with such a form, canonical transformations are its symmetries, and Hamiltonian dynamics is the flow the form assigns to each energy function. That viewpoint — Arnold's — is where Symplectic Group and Phase-Space Geometry goes next.