picture of mechanics




physics, mathematical physics, philosophy of physics

Surveys, textbooks and lecture notes

theory (physics), model (physics)

experiment, measurement, computable physics



Dynamics affects both observables and, dually, states; this is most well known in quantum mechanics but applies equally well to classical mechanics. The different “pictures” of mechanics differ in how the dynamics is explicitly formalized:

The pictures are named after those physicists (Werner Heisenberg, Erwin Schrödinger, and Paul Dirac) who first used or popularised these approaches to quantum physics.


With global time

Let us assume a global notion of time, say a fixed background? spacetime which is globally hyperbolic, so that it admits a foliation into Cauchy surfaces, and choose a time coordinate? for this foliation. The upshot of this is that each event occurs at a time tt, and conversely we can speak of space at any time tt (at least within certain bounds). Thus we may speak sensibly of either the state of the world at time tt or the value of some observable quantity at time tt.

Because this is a picture of dynamics, states or observables (as appropriate to the picture) will vary through time. We therefore have a time evolution? operator U(t,t)U(t,t') between any two times t,tt,t'; actually, we need consider only U(t)U(t,0)U(t) \coloneqq U(t,0), since U(t,t)=U(t)U(t) 1U(t,t') = U(t) \circ U(t')^{-1}.

In the Heisenberg picture, for each observable AA, we speak of AA only at some time tt, so our actual observables are of the form A(t)A(t). We write abstractly

A(t)=A(0)U(t) A(t) = A(0) \cdot U(t)

to show the evolution of the observable through time. However, when it comes to the state of the world, we speak of a single state ψ\psi that describes the world at all times.

In the Schrödinger picture, we instead speak of ψ(t)\psi(t), the state of the world at time tt. We write abstractly

ψ(t)=U(t)*ψ(0) \psi(t) = U(t) \ast \psi(0)

(the Schrödinger equation) to show the evolution of the state through time. However, when it comes to observables, we use only the observable AA across all times.

To see the connection between the two pictures, recall that an observable AA and a state ψ\psi together produce a probability distribution giving the probability that any given value of AA will be observed, given that the world is in state ψ\psi. (This is true throughout mechanics, although it is obscured in non-statistical classical mechanics, since the probability distributions produced by classical pure states are all delta measure?s.) Assuming that AA belongs to an appropriate algebra of observables and the probability measures are sufficiently nice, we may restrict attention to the expectation values A ψ\langle{A}\rangle_\psi of these distributions, since the entire distribution can be recovered from A n ψ\langle{A^n}\rangle_\psi as nn varies over natural numbers.

The connection between the two pictures is then given by

AU(t) ψ=A U(t)*ψ. \langle{A \cdot U(t)}\rangle_\psi = \langle{A}\rangle_{U(t) \ast \psi} .

It remains to say exactly what U(t)U(t) is and what the operations \cdot and *\ast are. Let us use the density matrix formulation of quantum statistical mechanics?, since classical and non-statistical mechanics may be recovered as special cases, by restricting (respectively) the allowed observables or states. In this case, both states and observables are given by linear operators on a Hilbert space HH, and we have

A ψ=tr(Aψ) \langle{A}\rangle_\psi = \tr(A \psi)

(using the trace operation). Each U(t)U(t) is a unitary operator on HH (since time evolution between Cauchy surfaces is a symmetry), and we have

AU(t)=U(t) 1AU(t) A \cdot U(t) = U(t)^{-1} A U(t)

(a right action?) and

U(t)*ψ=U(t)ψU(t) 1 U(t) \ast \psi = U(t) \psi U(t)^{-1}

(a left action?). We then have

AU(t) ψ=tr(U(t) 1AU(t)ψ)=tr(AU(t)ψU(t) 1)=A U(t)*ψ, \langle{A \cdot U(t)}\rangle_\psi = tr(U(t)^{-1} A U(t) \psi) = tr(A U(t) \psi U(t)^{-1}) = \langle{A}\rangle_{U(t) \ast \psi} ,

as desired, using the cyclic property of the trace?.

The time evolution operator U(t)U(t) is often derived from a Hamiltonian and the formula for A(t)A(t) or ψ(t)\psi(t) is further derived from a differential equation involving this Hamiltonian. However, this is unnecessary for the connection between the two pictures.

Without time

If spacetime is not globally hyperbolic, then there is no time coordinate tt, and none of the discussion above makes sense; or if we choose a coordinate tt and call it time regardless, then time evolution is not a symmetry and we do not have the operators U(t)U(t).

In this case, the Heisenberg picture still makes sense, even though we cannot expect to calculate A(t)A(t) from A(0)A(0) (if it even makes sense to discuss such things). This is easily seen in field theory, where the operators called AA above are really of the form A(x,y,z)A(x,y,z). Then the Heisenberg picture's A(t)A(t) is really A(x,y,z,t)A(x,y,z,t), or simply A(p)A(p) where pp indicates an event (a point in spacetime). So even if the coordinates x,y,z,tx,y,z,t do not make sense, still A(p)A(p) does; and even if the equations of physics cannot be thought of as describing evolution through time, still they can be thought of as describing the relationships between observables at different places in spacetime.

In contrast, the Schrödinger picture cannot be so treated. One may be led to the contrary impression by the quantum mechanics of a single particle without any internal structure (not even spin), in which case the Hilbert space of (pure quantum-mechanical) states is naturally identified with L 2( 3)L^2(\mathbb{R}^3) and the state ψ\psi is really ψ(x,y,z)\psi(x,y,z). In this case, the Schrödinger picture's ψ(t)\psi(t) is really ψ(x,y,z,t)\psi(x,y,z,t), that is ψ(p)\psi(p). However, this fails in classical or statistical mechanics; and even in non-statistical quantum mechanics, it breaks down if the particle has internal structure or there is more than one particle in the world. Then we see that the spacial coordinates x,y,zx,y,z generalise to the arbitrary coordinates of configuration space, while tt remains only tt, and there is no way to subsume it into a spacetime coordinate.


Historically, the terms ‘Schrödinger picture’ and ‘Heisenberg picture’ (at least) referred to more than what we discuss above; they referred to the entirety of the differences between Schrödinger's and Heisenberg's approaches to quantum mechanics.

For example, these terms included also Schrödinger's use of typically wave-like functions as pure states (and correspondingly operators in the higher-type-theoretic sense as observables) vs Heisenberg's use of infinite-dimensional matrices as observables (and correspondingly infinite sequences as pure states). This difference was rectified by von Neumann's application of Hilbert space to the problem, showing that (if one suitably restricts the allowed functions and sequences and also identifies equivalent functions a bit) both approaches used Hilbert space (what we would now call the infinite-dimensional separable Hilbert space) as the space of pure states.

This is entirely separate from the question of whether states or observables are taken to evolve with time. Still, there is this connection: Schrödinger evolved states, and his approach was called ‘wave mechanics’ after his representation for states, while Heisenberg evolved observables, and his approach was called ‘matrix mechanics’ after his representation for observables.

duality between algebra and geometry in physics:

Poisson algebraPoisson manifold
deformation quantizationgeometric quantization
algebra of observablesspace of states
Heisenberg pictureSchrödinger picture
higher algebrahigher geometry
Poisson n-algebran-plectic manifold
En-algebrashigher symplectic geometry
BD-BV quantizationhigher geometric quantization
factorization algebra of observablesextended quantum field theory
factorization homologycobordism representation


See for instance sections 7.19.1–3 in

  • Eberhard Zeidler, Quantum field theory. A bridge between mathematicians and physicists – volume I Springer (2009) (web).

To check conventions at least, see Wikipedia:

A note on how the Schrödinger picture in the form of extended FQFT on Lorentzian manifolds is related to the Heisenberg picture in the form of AQFT is in

Revised on September 24, 2014 08:49:46 by Toby Bartels (