State space

Lagrangian Mechanics gives us a foundational way to formulate the laws of motion. We write down an invariant quantity in a spacetime geometry, extremize it, and recover the physical path. The particle, we may say, has "learned" what the right trajectory is by solving an optimization problem. The global formulation is conceptually prior to the local formulation in the strict sense that the central objects of the differential formulation, the quantities that will become momentum and energy in the local theory, are derived from the global formulation.

"Hamiltonian Mechanics" is the name we give to the modern version of the local, differential perspective. Instead of selecting a whole history at once, it treats motion as the flow in time of one instantaneous state to the next. Hamiltonian mechanics is an elegant formulation of classical mechanics, but if elegance were all it provided, it would not belong here in our treatment of the conceptual foundations of physics. In fact, it does more than recast Lagrangian Mechanics in local form. It makes visible statistical and algebraic structure that is difficult to see in Lagrangian form and that later becomes essential to quantum mechanics. First, it reveals how statistical ensembles evolve in ways not visible from individual trajectories. Second, by encoding laws of global trajectories into the local geometry of infinitesimal state changes, it allows those laws to be expressed in an algebra of generators.

The Evolution of Ensembles in Physical Systems

At face value, the role of physical laws appears to be to predict individual trajectories from initial conditions. Yet in conservative systems, where energy is not dissipated into inaccessible microscopic degrees of freedom, the laws that govern the evolution of single states imply that the information encoded in a distribution of states is preserved. This is a categorically different kind of claim, not about one particle's initial conditions and its future, but about the collective behavior of an ensemble. The key to this emergent structure is that deterministic, reversible evolution keeps distinct initial states distinct. Once that is true, an ensemble can move through state space without losing the measure of how many states it contains.

Example: Preserved Information in a Permutation

We can get a feel for this kind of information-preserving evolution with a simple example of a system with a finite number of states. Take a deck of three cards labeled A, B, and C. The exact possible states are the six deck orders

ABC, ACB, BAC, BCA, CAB, CBA.

Now choose one specific permutation of these states, for example, move the top card to the bottom. Then

ABC -> BCA
BCA -> CAB
CAB -> ABC
ACB -> CBA
CBA -> BAC
BAC -> ACB

Now assign a probability distribution on these six exact states. For example,

P(ABC) = 1/2
P(BAC) = 1/3
P(CBA) = 1/6

with all other states assigned probability 0. After applying the permutation to all arrangements consistent with the distribution, the distribution becomes

P(BCA) = 1/2
P(ACB) = 1/3
P(BAC) = 1/6

with all other states still assigned probability 0. The same probability weights are still present. They have only been reassigned to different exact states.

A few structural facts are worth making explicit:

1. An initial arrangement is unique.
2. Under a given permutation rule, a new arrangement is dictated by the previous arrangement.
3. We can run the permutation in reverse and recover the original arrangement.
4. Two arrangements under the same number of permutations remain distinct.

These properties, which we will show exist in physical situations, are what we mean by "determinism" and "reversibility" and, following from these, by "distinguishability" of the ensemble under time evolution.

We can map the elements of the card deck to Hamiltonian mechanics. There, the relevant state space is the space of positions and momenta, or phase space. For many bodies, phase space is the product of each individual body's position/momentum space.

1. An arrangement of cards -> an instantaneous state
2. All possible such arrangements -> phase space
3. The count of distinct arrangements in a collection -> phase space area

The last of these elements is what we call a "measure" on the space. A measure has a role analogous to a metric. Where a metric measures separation, a measure assigns size to regions or collections. What matters here is whether that size is preserved as the system evolves. In the card deck, preservation is built in. The law is a permutation, so a collection of arrangements can be rearranged, but it cannot gain or lose members.

Example: Preserved Information in a Physical Ensemble

We see this same structure in conservative physical systems such as the ideal pendulum. Prepare many runs with slightly different initial angles and momenta. At the initial time, measurements across the ensemble reveal a distribution in phase space. We then let each member of the ensemble evolve under the same conservative law, and at some later time we again measure position and momentum across the ensemble. Those later measurements reveal a new distribution that has the same features as the original. Each exact initial state in the ensemble is carried to one exact later state, so the later distribution is the earlier one transported by a deterministic, reversible rule. In a dissipative system, nearby runs can truly be driven together, so distinctions among initial conditions are washed out, the extreme case being when all initial conditions end at rest. But in a conservative system, such distributions are not washed out. They are transported to new regions of state space.

The first animation below shows how nearby initial conditions can end in different regimes of motion. The second shows that even as an area patch is deformed to follow states in these different regimes, the area is preserved.

The pendulum is an analogue of the permutation example. There, the state space was a finite set, the natural measure was simply the count of arrangements, and the law was a permutation. Here, the state space is continuous. Its points are instantaneous angle/momentum states, its natural measure is area on state space, and its law of motion is a Hamiltonian flow. A permutation rearranges exact states without merging them. A Hamiltonian flow does the same for a continuum of exact states, and area in state space plays the role that counting measure played before. In conservative physical systems, this behavior is known as Liouville's theorem. Formally, if $\Phi_t$ is the Hamiltonian time-evolution map and $A$ is any measurable region of phase space, then

where $\mu$ is the natural measure on phase space.

Information as an Incompressible Fluid

In the examples above, we looked at spaces in which the invariance of a measure of arrangements was ensured by determinism and reversibility. A tactile example of this situation is incompressible fluid flow. Imagine marking out a connected patch on the surface and then tracking that same patch as it is carried by the flow. The patch may be stretched, folded, or sheared, but its area remains constant.

A planar flow is described by a velocity field. If a point at position $(x,y)$ has velocity $(u(x,y),v(x,y))$, then its motion is governed by

and incompressibility means

The two terms measure the rate at which a small material patch carried by the flow expands or contracts along the two coordinate directions. If their sum is zero, any expansion in one direction is exactly balanced by contraction in the other, so a small material patch does not locally change area. In two dimensions this condition has a powerful consequence. There exists a scalar function $\psi(x,y)$ such that

In two dimensions, incompressibility lets the velocity field be generated from this single scalar function in a cross-coupled form. Motion in the $x$ direction is determined by how the scalar function $\psi$ varies in the $y$ direction, and vice versa. Hamiltonian Mechanics shares this cross-coupled differential structure.

Embodied Information

We have described how systems preserve information as they evolve. Information cannot exist without some embodiment, be that in switches in a computer or molecules in a gas. To appreciate how physical systems embody information, we need to understand a bit about information theory itself.

How much information does a penny showing heads contain? A good way to think about this is as a question about surprise. If a coin were weighted so that heads always appeared, seeing heads would be completely unsurprising. If it showed heads 90% of the time, seeing heads would largely confirm expectations, and would therefore contain only a small amount of information. But if it came up tails, it would force a revision of expectations. This quantity is called "surprisal."

We also want information to count independent members of a distribution additively. If one fair coin toss resolves one binary uncertainty, then two independent fair coin tosses resolve two such uncertainties. We do not want to say that the second coin makes the information four times larger simply because the joint probability of two heads is $1/4$. This idea is second nature in computer engineering where the number of possible outcomes multiplies, but the number of independent binary distinctions adds. A register is "8-bit" even though it can realize $2^8$ possible bit strings.

These two requirements are met by defining the information content of an outcome using a base-2 logarithm:

When the probability of heads is 50%, this is exactly 1 bit. If the probability increases, it goes down. If it decreases, it goes up. Moreover, the information from 2 heads adds:

This quantity describes the information associated with one realized outcome. When we turn from one outcome to an ensemble, the natural scalar quantity is the average surprisal over the whole distribution. This is the entropy (or, more specifically, the Shannon entropy).

This same idea applies to an ensemble of initial conditions in a physical system. Rather than a discrete number of possibilities, there is a continuous set, so we must deal in probability densities rather than probabilities. With this change, the expression for the ensemble entropy becomes:

where $d\mu$ is the natural measure on the space of states. In this sense, a physical ensemble contains a quantifiable amount of information.

Liouville's theorem implies that this entropy is preserved in conservative evolution. But entropy is only a scalar summary of a distribution. Different densities can have the same entropy while assigning probability to different regions of state space. The stronger statement is that Hamiltonian evolution transports the density itself. If $\rho_0$ is the initial density, then the later density is the same probability structure carried forward by the flow. The information is embodied not merely in the number $H[\rho]$, but in the full distribution over possible states.

The information-theoretic implications of conservative systems are central to statistical mechanics. There, the exact microscopic state of a system is not practically knowable. A gas, for example, has too many microscopic degrees of freedom to track directly, and even a finite partition of phase space would leave an unfathomably large number of possible microscopic arrangements. Statistical mechanics works by replacing exact microscopic knowledge with distributions over possible states, and then asking how those distributions evolve. Macroscopic, thermodynamic quantities such as temperature, pressure, volume, and particle density are then understood statistically without requiring knowledge of specific states.

Yet the importance of the evolution of distributions in thermodynamics was but a harbinger of one of the handful of most paradigm-shifting theoretical advances in the history of science. In quantum mechanics, the state is itself a distribution, which encodes possible measurement outcomes. There is provably no definite state, the ignorance of which comprises the distribution, but only the distribution itself. We will return to this.

Phase Space

Phase space, or position/momentum space, is the arena in which Hamiltonian Mechanics takes place. To appreciate the Hamiltonian formulation, we need to understand the origin, geometry, and role of this space.

From Histories to Instantaneous States

The central objects in Hamiltonian mechanics, the state variables and the Hamiltonian function on those variables, are obtained from Lagrangian mechanics. We know that Lagrangian mechanics selects physical paths from path space by extremizing action. The task we must now perform is to extract the essential objects of a local theory of instantaneous state from the global Lagrangian theory.

As we know from Lagrangian mechanics, we map spacetime variables to physical histories through their contribution to the variation of the action. By integrating the action integral by parts, we can separate those contributions into the variation of the interior of the path and the variation of the endpoints. The former yields the Euler-Lagrange equations of motion. That is, the equations of motion must be consistent with the path taken once we know the endpoints. But what is the significance of the contribution to the action from displacing the endpoints? It does not answer, “How does the path taken from A to B change the action?” but rather, “How does an infinitesimal change in position, while holding time constant, change the action?” If we imagine chopping a path into smaller and smaller pieces, in a sense all that is left is a series of boundary terms, each corresponding to a successive instantaneous state. The boundary term records how the action changes when that state is infinitesimally displaced, and the coefficient of the displacement, or we might say the measure of the displacement’s effect on the action, will become the momentum.

To see this explicitly, start with the expression for the action:

Now vary the path $q(t)\to q(t)+\delta q(t)$. Because the Lagrangian depends on both $q$ and $\dot q$, varying the path also varies the velocity: $\dot q(t)\to \dot q(t)+\delta \dot q(t)$. The first-order change in $L$ is therefore just the multivariable chain rule applied to those two arguments. Then

The second term contains $\delta \dot q^i$ rather than $\delta q^i$. But the velocity variation is not independent. It is the time derivative of the path variation, $\delta \dot q^i = \frac{d}{dt}\delta q^i$. Since the path displacement $\delta q^i$ is the arbitrary object, integrate by parts to move that time derivative off the variation.

Substituting this back in gives

As advertised, the variation separates into the bulk term, which governs the equations of motion, and the boundary term, which governs the action's sensitivity to endpoint configuration displacement on a time slice.

The Definition of Momentum

The boundary term has the form $p_i\,\delta q^i$, with

This tells us what quantity pairs with an infinitesimal displacement of the state to measure its contribution to the change in action. That is, momentum, or more precisely conjugate momentum, is the gradient of the action with respect to configuration displacement at the boundary.

The functional form of the momentum is determined by the functional form of the Lagrangian. In a free system, in which only the kinetic term appears in the Lagrangian, $p$ arises inevitably from the theory's kinematics. In the Newtonian free-particle theory, the kinetic term is proportional to the square of the velocity, a fact that can be shown with some work we won't do here, to follow from Galilean symmetry.

In that setting the familiar formula $p=mv$ is inherited from the kinetic term. In a relativistic theory, by contrast, the free Lagrangian is constrained by Minkowski geometry. For a free particle, the action is proportional to proper time,

Writing this action in terms of coordinate time $t$, with $c=1$, gives

and therefore

The corresponding conjugate momentum is

From Momentum to the Symplectic Form

Thus far, the action has identified $q$ and $p$ as the correct paired variables for specifying instantaneous state.

This gives us what mathematicians call a one-form:

If we pass in the displacement $\delta q$,

We obtain the first-order change in the action.

The salient feature of Hamiltonian flows, their "incompressibility," requires a different type of mathematical object, one that measures an "amount of states," or an "area" on phase space.

Once the action has identified $q^i$ and $p_i$ as conjugate variables, the natural oriented area element on their space is the symplectic two-form. A two-form takes in two vectors and returns a number. The symplectic form is given the label $\omega$:

This area is the continuous analogue of counting states. One can see this by imagining phase space as a fine lattice. In that case, the count of lattice points is obviously proportional to the area enclosing them. In the continuous case, the area becomes the preserved measure without a discrete count of points. With more degrees of freedom, the corresponding volume measure is built from repeated products of $\omega$.

Once we have the symplectic form $\omega$, we can see the job of Hamiltonian mechanics as finding flows under which the area measured by the form is invariant. This is the geometric form of information preservation.

Defining Phase Space

We are now in a position to define phase space. It is the set of possible instantaneous states of a system described by generalized position coordinates and their conjugate momenta, as selected by the action. For one degree of freedom, a point in phase space is $(q,p)$. For many degrees of freedom, a point is $(q^1,\dots,q^n,p_1,\dots,p_n)$.

The space is equipped with the two-form

which measures the local amount of states and gives phase space its invariant symplectic structure.

Generating Time Evolution

Consider the contrast between phase-space animations and spacetime diagrams. In a spacetime diagram, time is an axis, motion is encoded into the worldline, and the history is contained in a static plot. In the phase-space picture, there is no physical motion, no history, until a point or region of phase space begins to flow. Our next task, then, if we wish to describe motion using phase space, is to characterize the flows of conservative systems. The Hamiltonian is the function that generates the physical evolution flow on phase space.

Generating Flows on Phase Space

We have already seen in our example of an incompressible fluid that any flow can be encoded by a generating function. Let's spell this out more carefully and put it in the arena of phase space.

Picture a contour map on the $q,p$ plane. The value of a function, $F(q,p)$, assigns a height to each point. This three-dimensional "hilly" picture can, as with topographic maps, be represented in two dimensions with contours, or level sets. The rule is that the denser the level sets are, or equivalently, the steeper the surface is, the faster the flow along that contour. This flow can then be represented as a vector field, that is, a phase-space velocity arrow at every point.

Enlarge Open file

We can convert $F$'s level sets into a vector field as follows:

These cross-coupled equations are the dynamical face of the geometric coupling that preserves the symplectic area form. Solving them for $q(t)$ and $p(t)$ gives the Hamiltonian flow.

A trivial example is $F=p$. Then

And integrating gives

That is, the flow generated by constant momentum is spatial translation.

Once $q$ and $p$ have been selected as conjugate variables, any smooth function $H(q,p)$ generates a flow on phase space. The special role of the Hamiltonian is that its flow is the physical time evolution of the system.

Preservation of the Symplectic Area

Now that we have a procedure for generating a flow from a smooth function, we can check that such a flow preserves the area form.

To see the preservation, ask what would have to happen for a small patch of phase space to gain or lose area. States would have to flow into the patch from a source or flow out of the patch through a sink. In vector-field language, local sources and sinks are measured by divergence.

In one dimension, if the velocity $\dot q$ gets larger as $q$ increases, then points slightly farther to the right move even farther right, so the flow spreads in the $q$-direction. The rate of that spreading is $\partial \dot q/\partial q$. Likewise, $\partial \dot p/\partial p$ measures spreading in the $p$-direction. Divergence adds these directional spreading rates.

For one degree of freedom, the phase-space velocity has two components, $\dot q$ and $\dot p$. Its divergence is

The first term asks whether nearby points separate in the $q$-direction as they move. The second asks whether nearby points separate in the $p$-direction. If the sum is zero, any stretching in one direction is balanced by squeezing in the other.

Now use the flow generated by $F$:

Substituting these into the divergence gives

For a smooth function, the order of these two derivatives does not matter:

So the two terms cancel, and the divergence vanishes:

Energy as the Hamiltonian

We have already seen the pairing of energy and time in relativity. The action for a free particle is proportional to the spacetime length of the worldline, and its terms pair energy with time and momentum with position:

Four-momentum is the covector that measures the contribution of a displacement in spacetime to action, and energy is the component that measures the contribution of a time displacement.

Enlarge Open file

Given that we are seeking a function to generate the flow of time, and that we know that time pairs with energy in spacetime, it is not unreasonable to guess that the Hamiltonian function is the energy of the system. Indeed, for conservative systems, this is exactly the case, as we now show.

For a relativistic free particle, the action is

In coordinate time,

so

with

The conjugate momentum is

The Hamiltonian is the Legendre transform:

Substituting gives

That is exactly the relativistic energy:

We may then say "energy generates time," a hard result, and a rich metaphor.

Hamiltonian Mechanics

We now know the geometry of phase space, we know we are looking for a function on it, and we know that function should be energy in accordance with relativity. What remains is to derive that function from the Lagrangian. This requires a change of variables, because the Lagrangian is written as

so it still treats velocity as an independent variable. The function we want must instead be written on phase space:

The Legendre Transform

The Legendre transform performs this change of variables by rewriting a function in terms of tangent-line data. Holding $q$ fixed, regard $L$ as a function of the velocity $\dot q$. Its slope in the velocity direction is

which is precisely the conjugate momentum. Trading velocity for this slope does not discard the rest of the tangent-line information. The remaining vertical information is encoded by $p_i\dot q^i-L$. The transformed function is

with $\dot q$ understood as the function of $(q,p)$ determined by the momentum equation above.

We can do a quick test that $H$ is indeed the total energy by looking at a free particle.

For a non-relativistic free particle,

The conjugate momentum is therefore

so

Substituting this into the Legendre transform gives

But this is exactly the kinetic energy of the particle. So in this simplest case, the Legendre transform gives the energy function directly.

Hamilton's Equations of Motion

The Legendre transform does more than rewrite energy in the desired variables. It produces the specific phase-space function whose generated flow reproduces the Euler-Lagrange motion. To see this, take the differential of $H$. In the coordinates we are using, this is the gradient-like object that will be turned 90 degrees to form the vector field generated by $H$:

Now use the momentum definition

The $d\dot q^i$ terms cancel. What remains is

At this stage the transform has done its essential job. The velocity direction no longer appears in the differential as an independent variation. The theory has been rewritten in phase-space terms.

To connect this with the original dynamics, use the Euler-Lagrange equations. The remaining Lagrangian derivative is not just a leftover term. It is the generalized force term. Since momentum was defined by $p_i=\partial L/\partial \dot q^i$, the Euler-Lagrange equation says that the derivative of $L$ in the $q^i$ direction is the time derivative of momentum:

Since $p_i = \partial L/\partial \dot q^i$, this becomes

This is where the dynamics selected by the action enters the phase-space expression. In ordinary mechanics, force is $dp/dt$; here the same idea appears as $\dot p_i$. Substituting this into the expression for $dH$ gives

But because $H$ is now a function of $(q,p)$, its differential is also

Match coefficients and, voila, we have the beautiful and useful Hamilton's equations of motion:

We can check this works using the example of a simple harmonic oscillator.

The total energy, and thus the Hamiltonian, is:

Hamilton's equations then give:

Using $p = m\dot q$, this becomes:

or

At this point we have the local differential formulation of mechanics in hand. The flow picture we have just developed can also be expressed as an algebra of functions on phase space, and that repackaging will make further structure visible.

The Algebra of Hamiltonian Mechanics

Given two functions $f$ and $g$ on phase space, we can ask how $f$ changes along the flow generated by $g$ or how $g$ changes along the flow generated by $f$.

As any two numbers can be combined via addition or multiplication to produce a new number, so any two functions on phase space can be combined to produce a new function. When the rule of combination reflects the flow/evolution relationship between functions, we get the algebra of Hamiltonian mechanics, or Poisson algebra. In that algebra, Hamilton's equations, conservation laws, and other structural relationships become visible as algebraic identities.

What We Mean by an Algebra

Before exploring Poisson algebra specifically, it is worth reminding ourselves what the word algebra means. An algebra begins with some class of objects and then specifies what operations may be performed on those objects and what properties those operations satisfy. Numbers, with operations of addition and multiplication satisfying the properties of associativity and commutativity, are the familiar example. Different algebras may involve different objects, operations, and properties, but these differences fall into a common taxonomy.

Taxonomy of Operations

Algebras are "closed," meaning that when allowed objects are combined, the result is again an allowed object of the same kind.

Taxonomy of Properties

Poisson Algebra as a Lie Algebra

With this framing in mind, we can now turn to Poisson algebra. Before discussing the specific operations and properties that form the Poisson algebra notation, let's first look at its geometric meaning. A Poisson algebra belongs to the broader class of Lie algebras, in which we represent a finite transformation on a manifold by an infinitesimal displacement tangent to the transformation.

This picture is captured schematically as a differential equation that represents the action of an infinitesimal generator on a finite transformation. The generator acts on the current state to produce its next infinitesimal change.

Here $G(t)$ is a one-parameter family of finite transformations, and $X$ is the infinitesimal generator, an element of the Lie algebra. The equation says that the finite transformation changes at each instant according to the infinitesimal generator. Repeating that infinitesimal step produces the finite transformation. In this sense, there is an exponential map from the generator to the transformation.

For matrix groups, this becomes

where $M(t)$ is the finite transformation matrix and $A$ is the infinitesimal generator matrix. Its solution is

The matrix form appears when a transformation can be represented by one linear rule acting on a vector space, such as a rotation, boost, or internal symmetry transformation. The same matrix acts everywhere, taking each vector to its transformed vector.

For a flow on phase space, the finite transformation is better understood as a map from phase space to itself generated by local instructions. The vector field assigns a direction to each point of phase space, and different points may be assigned different directions and speeds. It is therefore easier to write the same idea in terms of a moving point:

where $z_0$ is the initial phase-space point and $\phi_t$ is the finite transformation at parameter value $t$. The differential equation is

where $X_f$ is the vector field generated by the function $f$. In words, the phase-space point moves at each instant in the direction assigned by $X_f$ at the point's current location.

In the matrix case, the infinitesimal generator gives a single rule acting across the vector space. In the Hamiltonian case, the generator must determine a vector field on phase space. It assigns an infinitesimal direction to each point, with the whole field constrained by the symplectic form. The object that encodes this field is a function on phase space. Each smooth function $f$ determines a vector field $X_f$, and that vector field determines a flow of canonical transformations.

The chain of objects is

function -> vector field -> flow -> canonical transformations.

One may refer somewhat ambiguously to either the functions or the vector fields as the generator of the flow. Before writing the Poisson bracket in coordinates, we can say what it has to do geometrically. If $f$ and $g$ determine vector fields $X_f$ and $X_g$, those vector fields have their own Lie bracket, $[X_f,X_g]$. The Poisson bracket $\{f,g\}$ is the function whose vector field corresponds to that vector-field bracket:

This identity says that the Lie algebra of infinitesimal canonical transformations persists when one passes from vector fields to the functions that encode them.

The payoff is that one can work in the Lie algebra of infinitesimal generating functions rather than directly in the Lie group of finite transformations. The space of finite transformations can be pictured as a curved surface of transformations, where composition is the native operation. Infinitesimal generators live in a linear space tangent to the group at the identity, where objects can be added, scaled, and bracketed. This "flatness" is what makes an algebra possible here, and it is what makes insights and theorems visible in algebraic identities.

The Operations of the Algebra

The objects of Poisson algebra are smooth functions on phase space. The more elementary operations on this space are:

Hamiltonian mechanics adds an additional operation, the Poisson bracket:

With $g$ held fixed, $\{f,g\}$ is the change of $f$ when the state is moved along the vector field determined by $g$:

The result is another smooth function on phase space. It goes to zero when $f$ does not change along that flow.

The Properties of Poisson Algebra

The Poisson bracket is the algebraic formulation of the way Hamiltonian flows compose on phase space.

Bilinearity:

and likewise in the second slot.

Antisymmetry:

This sign flip is the algebraic form of oriented area. Swapping the inputs reverses the orientation.

Enlarge Open file

Jacobi Identity:

Operating on numbers moves a number line back and forth. It is visually obvious that no matter how one groups operations, the end placement of the number line is the same. When an algebra's objects are pictured as living in a plane, only cyclic permutations of the operands move the plane to the same location. Thus there is a kind of associativity, but it is more restrictive. This is the Jacobi identity:

Rather than saying that different nestings are equal, it says that the cyclic sum closes to zero.

Product Rule:

Poisson brackets act on products of functions as derivatives do in calculus. This is expected because the Poisson bracket differentiates one input along the flow of the other:

Physical Relationships Illuminated by Poisson Algebra

Once the Poisson bracket is in hand, physical relationships that previously had to be written as separate differential statements can be compressed into algebraic ones.

The Canonical Pairing of Position and Momentum

The are the identities that are meant when saying "momentum generates position translations" and "position generates momentum translations."

At this point the questioning reader may scratch their head. From relativity we expect momentum to be paired with position displacement, and from Noether’s theorem, we expect conserved momentum to follow from position symmetry. But what could it mean for position to generate momentum? Momentum symmetry? Conserved position? This is nonsense. The answer is that position and momentum are reciprocally paired as generators of canonical transformations, or phase space flows, not as generators of physical symmetries. A shift in position is a symmetry of a free physical system, a shift in momentum is not. The reciprocal generator relation is instead part of the phase-space structure that preserves information density. There is, nonetheless, a physical intuition behind it.

Imagine a downhill skier. The mountain's incline determines how their momentum changes, and that incline is a function of position. So $\dot p$ is determined by position. This example is illustrative because it realizes the idea of a potential "hill" as an actual hill. Whatever the details of the system, if we choose $p$ and $q$ so that $H = K(p) + V(q)$, then the more $H$ changes with respect to $q$, the greater the rate of change of momentum. This is what is meant, dynamically, by saying that $q$ generates $p$.

That distinction will matter later. Quantum mechanics realizes the position/momentum pairing in the representation of translations on waves. There, the classical relation $\{q,p\}=1$ becomes the commutator relation $[\hat q,\hat p]=i\hbar$.

Optional: Fourier phase and the quantum lift

While position/momentum pairing does not belong to the symmetries of classical physics, it does appear in the symmetry space of wavefunctions (if we associate wave number with momentum). The Fourier transform exchanges the two. Shifting in $x$ corresponds to multiplying by a $k$-phase, while multiplying by an $x$-phase shifts $k$. In that setting, $k$ generates translations in $x$, and $x$ generates translations in $k$. Quantum mechanics will turn this reciprocal Fourier symmetry into the canonical commutation relation, which underlies many aspects of quantum theory, including the structure of the Schrödinger equation and the Heisenberg uncertainty principle.

The similarity of wave function symmetry to symplectic geometry is not coincidental. In classical phase space, a position displacement and a momentum displacement can be represented by vectors such as $(a,0)$ and $(0,b)$. As we have seen, the symplectic form measures their oriented phase-space area:

$$ \omega((a,0),(0,b)) = ab. $$

Classically, these translations commute as motions of points. Moving a point by $a$ in position and then by $b$ in momentum reaches the same final point as doing the operations in the reverse order. But when the analogous transformations are represented on waves, the order matters. A position shift acts as

$$ (T_a\psi)(x)=\psi(x-a), $$

In wave-number space, a momentum shift is just a translation of the Fourier profile:

$$ \hat\psi(k) \mapsto \hat\psi(k-b). $$

But in position space, the same operation appears as multiplication by a position-dependent phase:

$$ (M_b\psi)(x)=e^{ibx}\psi(x). $$

Thus, multiplying a Fourier component by $e^{ibx}$ changes

$$ e^{ikx} \mapsto e^{i(k+b)x}. $$

Pop out video Open MP4

Applying them in opposite orders gives

$$ M_bT_a\psi(x)=e^{ibx}\psi(x-a), $$

while

$$ T_aM_b\psi(x)=e^{ib(x-a)}\psi(x-a). $$

Pop out video Open MP4

Thus

$$ M_bT_a = e^{iab}T_aM_b, $$

The phase factor is controlled by the same product $ab$ measured by the symplectic form:

$$ e^{iab}=e^{i\omega((a,0),(0,b))}. $$

With physical units restored, this becomes

$$ e^{\frac{i}{\hbar}\omega(v,w)}. $$

Thus the classical area pairing becomes, in wave mechanics, a phase relation. The symplectic form measures the failure of the corresponding wave transformations to commute. This is the Weyl-Heisenberg structure behind the canonical commutation relation. The Poisson bracket tells us classically that $q$ and $p$ are conjugate, while Fourier duality shows how that conjugacy acts on waves. Quantum mechanics then, as we will see, turns it into the operator identity:

$$ [\hat q,\hat p]=i\hbar. $$

Enlarge Open file

Hamiltonian Dynamics

The master dynamical statement is that, for any smooth function $f$ on phase space,

assuming $f$ has no explicit time dependence. In this form, time evolution itself becomes a bracket relation. Hamilton's equations are the special cases obtained by taking $f=q$ and $f=p$:

Conservation and Noether's Theorem (Forward and Backward)

Conservation laws also become algebraic statements. If

then $\dot p=0$, so momentum is conserved along the Hamiltonian evolution. More generally, if

then the quantity $f$ is conserved in time.

Because the bracket is antisymmetric, the same vanishing relation may be read in reverse:

Enlarge Open file

This is the compact meeting point of the two Noether directions. Read one way, it says that $f$ is conserved under the time evolution generated by $H$. Read the other way, it says that the transformation generated by $f$ leaves $H$ unchanged. Conservation and symmetry thus appear as two readings of the same algebraic relation.

Looking Ahead to Quantum Mechanics

This algebraic viewpoint provides a bridge from Hamiltonian mechanics into quantum mechanics. There, the classical geometry of states as points moving through phase space no longer carries over in the same direct form, and much of the new theory remains to be understood on its own terms. What does carry over cleanly is the algebraic structure of observables, generators, and their bracket relations. By replacing functions with operators and Poisson brackets with commutators, one can lift a large part of Hamiltonian structure before learning the geometry quantum theory requires.