3175 words
16 minutes
Hermitian vs. Self-Adjoint: The Momentum Operator

Note: This blog post is based on my own term paper for MIT 8.06x (Applications of Quantum Mechanics). You can find the original PDF here: Case Study of the Momentum Operator.

Why this post exists#

Most quantum mechanics textbooks say that observables correspond to “Hermitian operators”. Charitably, this is a half-truth, and the missing half is exactly the one that gets you into trouble as soon as you try to define a momentum operator on anything more interesting than Rn\mathbb{R}^n.

The issue is that “Hermitian” in the physics sense usually means symmetric, while what observables really need to be is self-adjoint, and in infinite dimensions these are not the same thing. The gap between them is invisible in finite-dimensional linear algebra, where every linear operator is bounded, so every densely defined symmetric operator is automatically self-adjoint. In quantum mechanics, where the Hilbert spaces are typically L2(Rn)L^2(\mathbb{R}^n) and the operators we care about are unbounded, the distinction cannot be avoided.

This post is a case study. We go through the definitions carefully (operator, adjoint, symmetric, self-adjoint, essentially self-adjoint) and then apply them to the momentum operator P^=−iℏ ∂x\hat{P} = -i\hbar\,\partial_x in two settings.

  1. A compact interval [0,2π][0, 2\pi] with hard-wall (Dirichlet) boundary conditions.
  2. A circle, i.e. the same interval with periodic boundary conditions.

In short, on the circle the momentum operator is essentially self-adjoint, and there is a unique sensible way to “promote” it to a true observable. On the interval with Dirichlet boundary conditions it is not essentially self-adjoint, and the question of which self-adjoint operator deserves to be called “the momentum on [0,2π][0,2\pi]” has no unique answer. There is a one-parameter family of options, and one has to choose. This is the kind of subtlety that gets lost when −iℏ ∂x-i\hbar\,\partial_x is treated as if it were a finite matrix.

I will assume a working knowledge of Hilbert spaces and basic functional analysis (bounded operators, dense subspaces, L2L^2), and set ℏ=1\hbar = 1 throughout.


1. Operators, bounded and unbounded#

An operator between two normed spaces AA and BB is just a linear map T:A→BT: A \to B. Nothing special so far, the interesting part is the topology.

In finite dimensions every linear operator is automatically continuous, because all norms on a finite-dimensional vector space are equivalent. This is false in infinite dimensions, and the failure is not a mathematical curiosity. It is the reason quantum mechanics needs a more careful framework than ordinary linear algebra.

Bounded operators#

Let (V,∥⋅∥V)(V, \|\cdot\|_V) be a normed space and (W,∥⋅∥W)(W, \|\cdot\|_W) a Banach space. A linear operator A:V→WA: V \to W is bounded if

sup⁡f∈V∖{0}∥Af∥W∥f∥V<∞,\sup_{f \in V \setminus \{0\}} \frac{\|Af\|_W}{\|f\|_V} < \infty,

or equivalently if there is a constant C≥0C \geq 0 such that ∥Ax∥W≤C∥x∥V\|Ax\|_W \leq C\|x\|_V for all x∈Vx \in V.

For linear operators, “bounded” and “continuous” mean the same thing (this is a standard lemma). We write B(H)\mathcal{B}(\mathcal{H}) for the space of bounded operators H→H\mathcal{H} \to \mathcal{H}.

Unbounded operators#

A linear operator that fails the boundedness condition is unbounded. The two most important operators in quantum mechanics, position and momentum, are both unbounded on L2(Rn)L^2(\mathbb{R}^n),

  • Position: x^ψ(x)=xψ(x)\hat{x}\psi(x) = x\psi(x).
  • Momentum: p^ψ(x)=−i ∂xψ(x)\hat{p}\psi(x) = -i\,\partial_x\psi(x).

Why are they unbounded? For position, one can find L2L^2 functions concentrated arbitrarily far from the origin, where ∥xψ∥/∥ψ∥\|x\psi\|/\|\psi\| blows up. For momentum, differentiation amplifies high-frequency components without bound.

A more elementary way to see that the derivative is unbounded is to define

D:C1[0,1]→C0[0,1],f↦f′.D: C^1[0,1] \to C^0[0,1], \qquad f \mapsto f'.

and take fn(x)=sin⁡(nπx)f_n(x) = \sin(n\pi x). Then ∥fn∥∞=1\|f_n\|_\infty = 1 but ∥fn′∥∞=nπ→∞\|f_n'\|_\infty = n\pi \to \infty, so the ratio is unbounded.

Important consequence. An unbounded operator cannot be defined on the whole Hilbert space (at least not as a closed linear operator). An everywhere-defined closed operator is automatically bounded (the closed graph theorem), and the special case for symmetric operators is the Hellinger–Toeplitz theorem. So an unbounded operator always lives on a proper subspace, its domain, and specifying the domain is part of specifying the operator. Two operators with the same formal expression but different domains are different operators. If there is one thing to take away from this section, it is this.


2. Adjoints, symmetric and self-adjoint#

Densely defined operators#

Let H\mathcal{H} be a Hilbert space. A linear operator T:DT→HT: \mathcal{D}_T \to \mathcal{H} is densely defined if its domain DT\mathcal{D}_T is dense in H\mathcal{H}, i.e. every vector in H\mathcal{H} can be approximated arbitrarily well by vectors in DT\mathcal{D}_T.

Density is what allows us to define the adjoint at all. Without it, the adjoint is not even well-defined as a single-valued operator.

The adjoint#

Let T:DT→HT: \mathcal{D}_T \to \mathcal{H} be densely defined. The adjoint T∗T^* is the operator with domain

DT∗:={ψ∈H ∣ ∃ η∈H such that ⟨ψ∣Tφ⟩=⟨η∣φ⟩ for all φ∈DT},\mathcal{D}_{T^*} := \big\{ \psi \in \mathcal{H} \,\big|\, \exists\,\eta \in \mathcal{H}\ \text{such that}\ \langle\psi | T\varphi\rangle = \langle\eta|\varphi\rangle \text{ for all } \varphi \in \mathcal{D}_T \big\},

acting as T∗ψ:=ηT^*\psi := \eta.

T∗T^* is well-defined. If both η\eta and η~\tilde\eta satisfy ⟨ψ∣Tφ⟩=⟨η∣φ⟩=⟨η~∣φ⟩\langle\psi|T\varphi\rangle = \langle\eta|\varphi\rangle = \langle\tilde\eta|\varphi\rangle for all φ∈DT\varphi \in \mathcal{D}_T, then ⟨η−η~∣φ⟩=0\langle\eta - \tilde\eta | \varphi\rangle = 0 for all φ\varphi in a dense set, which forces η=η~\eta = \tilde\eta. (This is exactly where density is used.)

A small but useful proposition is ker⁡(T∗)=ran⁡(T)⊥\ker(T^*) = \operatorname{ran}(T)^\perp. To see it, ψ∈ker⁡(T∗)  ⟺  T∗ψ=0  ⟺  ⟨ψ∣Tφ⟩=⟨0∣φ⟩=0\psi \in \ker(T^*) \iff T^*\psi = 0 \iff \langle\psi|T\varphi\rangle = \langle 0|\varphi\rangle = 0 for all φ∈DT  ⟺  ψ⊥ran⁡(T)\varphi \in \mathcal{D}_T \iff \psi \perp \operatorname{ran}(T).

Extensions and an inclusion lemma#

We say T~\tilde T is an extension of TT, written T⊆T~T \subseteq \tilde T, if DT⊆DT~\mathcal{D}_T \subseteq \mathcal{D}_{\tilde T} and T~φ=Tφ\tilde T\varphi = T\varphi for all φ∈DT\varphi \in \mathcal{D}_T.

Lemma. If T⊆T~T \subseteq \tilde T are both densely defined, then T~∗⊆T∗\tilde T^* \subseteq T^*.

Proof. Let ψ∈DT~∗\psi \in \mathcal{D}_{\tilde T^*}, so there is an η∈H\eta \in \mathcal{H} with ⟨ψ∣T~β⟩=⟨η∣β⟩\langle\psi|\tilde T\beta\rangle = \langle\eta|\beta\rangle for all β∈DT~\beta \in \mathcal{D}_{\tilde T}. Since DT⊆DT~\mathcal{D}_T \subseteq \mathcal{D}_{\tilde T} and T~=T\tilde T = T on DT\mathcal{D}_T, this gives ⟨ψ∣Tα⟩=⟨η∣α⟩\langle\psi|T\alpha\rangle = \langle\eta|\alpha\rangle for all α∈DT\alpha \in \mathcal{D}_T. Hence ψ∈DT∗\psi \in \mathcal{D}_{T^*} and T∗ψ=η=T~∗ψT^*\psi = \eta = \tilde T^*\psi. □\square

Note that the direction is reversed, i.e. a bigger operator has a smaller adjoint. We will use this repeatedly.

Symmetric vs. self-adjoint#

A densely defined operator TT is symmetric if

⟨α∣Tβ⟩=⟨Tα∣β⟩for all α,β∈DT.\langle \alpha | T\beta\rangle = \langle T\alpha | \beta \rangle \qquad \text{for all } \alpha, \beta \in \mathcal{D}_T.

It is self-adjoint if T=T∗T = T^*, which means that both the actions and the domains agree,

  1. DT=DT∗\mathcal{D}_T = \mathcal{D}_{T^*},
  2. Tφ=T∗φT\varphi = T^*\varphi for all φ∈DT\varphi \in \mathcal{D}_T.

A word of warning about language. In physics, “Hermitian” is used inconsistently, sometimes as a synonym for symmetric and sometimes as a synonym for self-adjoint. Statements like “observables correspond to Hermitian operators” gloss over a real distinction that only disappears in finite dimensions, so I will avoid the word entirely.

If TT is symmetric, then T⊆T∗T \subseteq T^*. (For ψ∈DT\psi \in \mathcal{D}_T, set η:=Tψ\eta := T\psi. Symmetry gives ⟨ψ∣Tα⟩=⟨η∣α⟩\langle\psi|T\alpha\rangle = \langle\eta|\alpha\rangle for all α∈DT\alpha \in \mathcal{D}_T, so ψ∈DT∗\psi \in \mathcal{D}_{T^*} with T∗ψ=TψT^*\psi = T\psi.) The whole question is then whether this inclusion is an equality.

A maximality property#

Self-adjoint operators have no proper symmetric extensions. Concretely, if TT is self-adjoint, T~\tilde T is symmetric and T⊆T~T \subseteq \tilde T, then T=T~T = \tilde T.

Proof. From T⊆T~T \subseteq \tilde T the lemma gives T~∗⊆T∗\tilde T^* \subseteq T^*. Symmetry of T~\tilde T means T~⊆T~∗\tilde T \subseteq \tilde T^*, so T~⊆T~∗⊆T∗=T\tilde T \subseteq \tilde T^* \subseteq T^* = T. Together with T⊆T~T \subseteq \tilde T this gives T=T~T = \tilde T. □\square

This is why self-adjointness is the right condition for observables. The spectral theorem guarantees that self-adjoint operators have real spectra and a sensible functional calculus, which symmetric operators in general do not. The maximality property says that a self-adjoint operator is “as big as it can be”, i.e. there is no room to extend it further and keep it symmetric.


3. Closability, closure, essential self-adjointness#

We need one more layer. Even if TT is not self-adjoint, it may have a unique self-adjoint extension hiding inside T∗T^*, and this is what essential self-adjointness captures.

Closures#

A densely defined operator TT is closable if its adjoint T∗T^* is itself densely defined. Equivalently (and more usefully), TT has a smallest closed extension, the closure T‾\overline T, which satisfies T‾=T∗∗\overline T = T^{**}.

Symmetric operators are always closable. If TT is symmetric, then T⊆T∗T \subseteq T^* implies DT⊆DT∗\mathcal{D}_T \subseteq \mathcal{D}_{T^*}. Since DT\mathcal{D}_T is dense, so is DT∗\mathcal{D}_{T^*}, and TT is closable. So for symmetric operators we always have T‾=T∗∗\overline T = T^{**} available.

For symmetric TT, the closure sits between TT and T∗T^*,

T⊆T∗∗⊆T∗.T \subseteq T^{**} \subseteq T^*.

Essentially self-adjoint operators#

A symmetric operator TT is essentially self-adjoint (e.s.a.) if its closure T‾=T∗∗\overline T = T^{**} is self-adjoint.

This is weaker than self-adjointness. A self-adjoint operator is automatically e.s.a., since T=T∗T = T^* gives, after taking adjoints, T∗=T∗∗T^* = T^{**}, so T=T∗∗T = T^{**} and T‾=T\overline T = T is self-adjoint.

Theorem. If TT is essentially self-adjoint, then T‾\overline T is the unique self-adjoint extension of TT.

Proof. T‾\overline T is a self-adjoint extension by hypothesis. For uniqueness, let SS be any other self-adjoint extension, T⊆ST \subseteq S. Self-adjoint operators are closed (since S=S∗S = S^* and adjoints are always closed), so T‾⊆S\overline T \subseteq S, as the closure is the smallest closed extension. But T‾\overline T is itself self-adjoint, and a self-adjoint operator has no proper symmetric extensions, so T‾=S\overline T = S. □\square

The practical upshot. To turn a symmetric operator into a genuine observable, we do not need it to be self-adjoint already. It is enough that it is essentially self-adjoint. The closure T‾\overline T is then the canonical observable, and it is unique.

When essential self-adjointness fails, life gets harder. The symmetric operator may then have many self-adjoint extensions (or none at all), and choosing one becomes a piece of physical input. The tool for counting and parametrizing these extensions is von Neumann’s deficiency index theory. It is beyond the scope of this post, but it comes up explicitly in the interval case below.


4. The momentum operator: setup#

Now to the main part. We define the momentum operator on the jj-th coordinate as

P^j:DP→L2,ψ↦−i ∂jψ,\hat{P}_j: \mathcal{D}_P \to L^2,\qquad \psi \mapsto -i\,\partial_j\psi,

where the domain DP\mathcal{D}_P still has to be specified. We work on H=L2([0,2π])\mathcal{H} = L^2([0, 2\pi]) in one dimension and consider two natural-looking domains,

  • Interval (Dirichlet): DPint={ψ∈C1([0,2π])∣ψ(0)=ψ(2π)=0}\mathcal{D}_P^{\text{int}} = \{\psi \in C^1([0, 2\pi]) \mid \psi(0) = \psi(2\pi) = 0\}.
  • Circle (periodic): DPcirc={ψ∈C1([0,2π])∣ψ(0)=ψ(2π)}\mathcal{D}_P^{\text{circ}} = \{\psi \in C^1([0, 2\pi]) \mid \psi(0) = \psi(2\pi)\}.

The Dirichlet condition is strictly stronger than the periodic one (forcing both endpoints to vanish implies periodicity, but not the other way round), so as operators P^jint⊊P^jcirc\hat{P}_j^{\text{int}} \subsetneq \hat{P}_j^{\text{circ}}, i.e. the circle operator is an extension of the interval operator.

We also need a couple of facts about function spaces. The chain of inclusions

C1([a,b])⊊H1([a,b])⊆AC([a,b])C^1([a,b]) \subsetneq H^1([a,b]) \subseteq AC([a,b])

relates classically differentiable functions (C1C^1) to absolutely continuous functions (ACAC) and to the Sobolev space H1H^1, where

  • ψ∈AC([a,b])\psi \in AC([a,b]) if there is a Lebesgue-integrable ρ\rho such that ψ(x)=ψ(a)+∫axρ(y) dy\psi(x) = \psi(a) + \int_a^x \rho(y)\,dy, in which case ρ=ψ′\rho = \psi' almost everywhere,
  • H1([a,b])={ψ∈AC([a,b])∣ψ′∈L2}H^1([a,b]) = \{\psi \in AC([a,b]) \mid \psi' \in L^2\}.

The point of H1H^1 is that it is the natural domain for ”ψ\psi has an L2L^2 derivative”. It lets us apply −i∂x-i\partial_x and stay in L2L^2, without requiring classical differentiability everywhere.


5. Momentum on the interval [0,2π][0, 2\pi]#

Take P^jint\hat{P}_j^{\text{int}} with domain DP={ψ∈C1([0,2π])∣ψ(0)=ψ(2π)=0}\mathcal{D}_P = \{\psi \in C^1([0, 2\pi]) \mid \psi(0) = \psi(2\pi) = 0\}.

Step 1. P^jint\hat{P}_j^{\text{int}} is symmetric#

For ψ,φ∈DP\psi, \varphi \in \mathcal{D}_P we integrate by parts,

⟨ψ∣P^jφ⟩=∫02πψ(x)‾ (−iφ′(x)) dx=−i[ψ‾φ]02π+i∫02πψ′(x)‾ φ(x) dx.\langle \psi | \hat{P}_j \varphi\rangle = \int_0^{2\pi} \overline{\psi(x)}\,\big(-i\varphi'(x)\big)\,dx = -i\big[\overline{\psi}\varphi\big]_0^{2\pi} + i \int_0^{2\pi} \overline{\psi'(x)}\,\varphi(x)\,dx.

The boundary term vanishes because ψ(0)=ψ(2π)=0\psi(0) = \psi(2\pi) = 0, and the remaining integral is

i∫02πψ′‾ φ dx=∫02π(−iψ′)‾ φ dx=⟨P^jψ∣φ⟩.i \int_0^{2\pi} \overline{\psi'}\,\varphi\,dx = \int_0^{2\pi} \overline{(-i\psi')}\,\varphi\,dx = \langle \hat{P}_j\psi | \varphi\rangle.

So P^jint\hat{P}_j^{\text{int}} is symmetric. ✓

Step 2. The adjoint (P^jint)∗(\hat{P}_j^{\text{int}})^* has a bigger domain#

Now we ask for which ψ∈L2\psi \in L^2 there is an η∈L2\eta \in L^2 with

∫02πψ‾ (−iφ′) dx=∫02πη‾ φ dxfor all φ∈DP?(⋆)\int_0^{2\pi} \overline{\psi}\,(-i\varphi')\,dx = \int_0^{2\pi} \overline\eta\,\varphi\,dx \qquad \text{for all } \varphi \in \mathcal{D}_P? \tag{$\star$}

The strategy is to pick any N∈AC([0,2π])N \in AC([0,2\pi]) with N′=ηN' = \eta a.e. (such an NN exists, since every L2L^2 function on a compact interval has an absolutely continuous antiderivative). Substituting η=N′\eta = N' and integrating the right-hand side by parts,

∫02πN′‾ φ dx=[N‾φ]02π−∫02πN‾ φ′ dx,\int_0^{2\pi} \overline{N'}\,\varphi\,dx = \big[\overline{N}\varphi\big]_0^{2\pi} - \int_0^{2\pi} \overline N\,\varphi'\,dx,

where the boundary term vanishes because φ∈DP\varphi \in \mathcal{D}_P vanishes at the endpoints. So (⋆)(\star) becomes

∫02π(ψ−iN)‾ φ′ dx=0for all φ∈DP.\int_0^{2\pi} \overline{(\psi - iN)}\,\varphi'\,dx = 0 \qquad \text{for all } \varphi \in \mathcal{D}_P.

or, equivalently, ψ−iN⊥{φ′∣φ∈DP}\psi - iN \perp \{\varphi' \mid \varphi \in \mathcal{D}_P\} in L2L^2.

Next we identify this orthogonal complement. The set {φ′∣φ∈DP}\{\varphi' \mid \varphi \in \mathcal{D}_P\} consists of exactly those continuous functions ξ\xi on [0,2π][0, 2\pi] with ∫02πξ(x) dx=0\int_0^{2\pi} \xi(x)\,dx = 0, i.e. ξ⊥1\xi \perp \mathbf{1}, where 1\mathbf{1} is the constant function. (In one direction, if ξ=φ′\xi = \varphi' with φ(0)=φ(2π)=0\varphi(0) = \varphi(2\pi) = 0, then ∫02πξ dx=φ(2π)−φ(0)=0\int_0^{2\pi}\xi\,dx = \varphi(2\pi) - \varphi(0) = 0. In the other, given such a ξ\xi, the antiderivative φ(x)=∫0xξ(y) dy\varphi(x) = \int_0^x \xi(y)\,dy vanishes at both endpoints and lies in C1C^1.)

So the L2L^2 closure of {φ′∣φ∈DP}\{\varphi' \mid \varphi \in \mathcal{D}_P\} is {1}⊥\{\mathbf{1}\}^\perp, and its orthogonal complement is {1}⊥⊥=span⁡{1}‾=C⋅1\{\mathbf{1}\}^{\perp\perp} = \overline{\operatorname{span}\{\mathbf{1}\}} = \mathbb{C}\cdot\mathbf{1}, the constant functions. Hence

ψ−iN=constant⟹ψ=constant+iN.\psi - iN = \text{constant} \quad\Longrightarrow\quad \psi = \text{constant} + iN.

Since N∈ACN \in AC, this gives ψ∈AC([0,2π])\psi \in AC([0, 2\pi]), with no boundary conditions at all. The requirement P^j∗ψ=η=N′=−iψ′∈L2\hat{P}_j^*\psi = \eta = N' = -i\psi' \in L^2 then forces ψ′∈L2\psi' \in L^2, i.e. ψ∈H1([0,2π])\psi \in H^1([0, 2\pi]). Conversely, every H1H^1 function arises this way, so

 D(P^jint)∗=H1([0,2π]),(P^jint)∗ψ=−iψ′. \boxed{\ \mathcal{D}_{(\hat{P}_j^{\text{int}})^*} = H^1([0, 2\pi]),\qquad (\hat{P}_j^{\text{int}})^*\psi = -i\psi'.\ }

The adjoint is the same differential expression on a much larger domain, H1H^1 functions with no boundary conditions, compared to C1C^1 functions that vanish at both endpoints.

In particular P^jint⊊(P^jint)∗\hat{P}_j^{\text{int}} \subsetneq (\hat{P}_j^{\text{int}})^*, so P^jint\hat{P}_j^{\text{int}} is not self-adjoint.

Step 3. The closure P^j∗∗\hat{P}_j^{**} is also too small#

Maybe the closure saves us? We compute DP∗∗\mathcal{D}_{P^{**}} directly.

ψ∈DP∗∗\psi \in \mathcal{D}_{P^{**}} means that there is a ζ∈L2\zeta \in L^2 with ⟨ψ∣P^j∗φ⟩=⟨ζ∣φ⟩\langle\psi | \hat P_j^* \varphi\rangle = \langle \zeta | \varphi\rangle for all φ∈DP∗=H1\varphi \in \mathcal{D}_{P^*} = H^1. Since P^j∗∗⊆P^j∗\hat P_j^{**} \subseteq \hat P_j^* and both act as −i∂x-i\partial_x, we have ζ=−iψ′\zeta = -i\psi', and the question is whether the boundary terms vanish.

For ψ∈H1\psi \in H^1 (which we already know is necessary, since DP∗∗⊆DP∗=H1\mathcal{D}_{P^{**}} \subseteq \mathcal{D}_{P^*} = H^1) and φ∈H1\varphi \in H^1, integrating by parts gives

⟨ψ∣P^j∗φ⟩−⟨P^j∗ψ∣φ⟩=−i[ψ‾ φ]02π.\langle\psi | \hat P_j^*\varphi\rangle - \langle \hat P_j^*\psi | \varphi\rangle = -i\big[\overline\psi\,\varphi\big]_0^{2\pi}.

This has to vanish for all φ∈H1\varphi \in H^1, and H1H^1 functions can take arbitrary, independent values at 00 and 2π2\pi, so ψ(0)=0\psi(0) = 0 and ψ(2π)=0\psi(2\pi) = 0. Therefore

DP∗∗={ψ∈H1([0,2π])∣ψ(0)=ψ(2π)=0},\mathcal{D}_{P^{**}} = \{\psi \in H^1([0, 2\pi]) \mid \psi(0) = \psi(2\pi) = 0\},

which is strictly smaller than DP∗=H1([0,2π])\mathcal{D}_{P^*} = H^1([0, 2\pi]). Hence P^j∗∗⊊P^j∗\hat{P}_j^{**} \subsetneq \hat{P}_j^*, and P^jint\hat{P}_j^{\text{int}} is not essentially self-adjoint.

What is really going on#

This is not a defect of our choice of starting domain. It reflects a real physical ambiguity. There is in fact a one-parameter family of self-adjoint extensions of P^jint\hat{P}_j^{\text{int}}, labelled by θ∈[0,2π)\theta \in [0, 2\pi), with domains

Dθ={ψ∈H1([0,2π])∣ψ(2π)=eiθψ(0)}.\mathcal{D}_\theta = \{\psi \in H^1([0, 2\pi]) \mid \psi(2\pi) = e^{i\theta}\psi(0)\}.

Different values of θ\theta correspond to physically different systems (for instance a particle on an interval threaded by a magnetic flux, or a “twisted” boundary condition). The deficiency-index analysis confirms that this is the whole story. The deficiency indices of P^jint\hat P_j^{\text{int}} are (1,1)(1, 1), which gives exactly a U(1)U(1) worth of self-adjoint extensions.

The lesson is that when an operator is symmetric but not essentially self-adjoint, the formal expression alone does not tell you which observable to use. You have to supply a boundary condition, and that choice is part of defining the system.


6. Momentum on the circle#

Now we repeat the analysis with the milder boundary condition ψ(0)=ψ(2π)\psi(0) = \psi(2\pi), without requiring the common value to be zero.

Step 1. Symmetric#

For ψ,φ\psi, \varphi with ψ(0)=ψ(2π)\psi(0) = \psi(2\pi) and φ(0)=φ(2π)\varphi(0) = \varphi(2\pi), the boundary term from integrating by parts is

−i[ψ‾φ]02π=−i(ψ(2π)‾φ(2π)−ψ(0)‾φ(0))=−iψ(0)‾(φ(2π)−φ(0))=0,-i\big[\overline\psi\varphi\big]_0^{2\pi} = -i\big(\overline{\psi(2\pi)}\varphi(2\pi) - \overline{\psi(0)}\varphi(0)\big) = -i\overline{\psi(0)}\big(\varphi(2\pi) - \varphi(0)\big) = 0,

where we used both boundary conditions. So P^jcirc\hat P_j^{\text{circ}} is symmetric. ✓

Step 2. The adjoint#

We can shorten the computation using the inclusion P^jint⊊P^jcirc\hat{P}_j^{\text{int}} \subsetneq \hat{P}_j^{\text{circ}}. By the inclusion lemma, (P^jcirc)∗⊆(P^jint)∗=P^j∗∣H1(\hat{P}_j^{\text{circ}})^* \subseteq (\hat{P}_j^{\text{int}})^* = \hat{P}_j^*|_{H^1}, so D(Pcirc)∗⊆H1([0,2π])\mathcal{D}_{(P^{\text{circ}})^*} \subseteq H^1([0, 2\pi]), and the action is still −i∂x-i\partial_x.

To fix the boundary conditions, ψ∈D(Pcirc)∗\psi \in \mathcal{D}_{(P^{\text{circ}})^*} requires the boundary term to vanish for all φ\varphi with φ(0)=φ(2π)\varphi(0) = \varphi(2\pi),

−i[ψ‾φ]02π=−i(ψ(2π)−ψ(0))‾ φ(0)=0for all such φ.-i\big[\overline\psi\varphi\big]_0^{2\pi} = -i\overline{\big(\psi(2\pi) - \psi(0)\big)}\,\varphi(0) = 0 \qquad \text{for all such } \varphi.

Since φ(0)\varphi(0) is arbitrary, this forces ψ(2π)=ψ(0)\psi(2\pi) = \psi(0), so

 D(P^jcirc)∗={ψ∈H1([0,2π])∣ψ(0)=ψ(2π)}. \boxed{\ \mathcal{D}_{(\hat{P}_j^{\text{circ}})^*} = \{\psi \in H^1([0, 2\pi]) \mid \psi(0) = \psi(2\pi)\}.\ }

i.e. the adjoint inherits the same periodic boundary condition, only on a bigger function space.

Step 3. Not self-adjoint, but…#

Comparing the two,

DPcirc={ψ∈C1∣ψ(0)=ψ(2π)}⊊{ψ∈H1∣ψ(0)=ψ(2π)}=D(Pcirc)∗.\mathcal{D}_{P^{\text{circ}}} = \{\psi \in C^1 \mid \psi(0) = \psi(2\pi)\} \subsetneq \{\psi \in H^1 \mid \psi(0) = \psi(2\pi)\} = \mathcal{D}_{(P^{\text{circ}})^*}.

so P^jcirc\hat{P}_j^{\text{circ}} is not self-adjoint either, but only because of the regularity gap C1⊊H1C^1 \subsetneq H^1, and not because of any mismatch in boundary conditions.

Step 4. Essentially self-adjoint!#

Now the closure. For ψ∈DP∗∗\psi \in \mathcal{D}_{P^{**}} and φ∈DP∗={H1 with periodic BC}\varphi \in \mathcal{D}_{P^*} = \{H^1\text{ with periodic BC}\}, the boundary term is

−i[ψ‾φ]02π=−i(ψ(2π)−ψ(0))‾ φ(0)-i\big[\overline\psi\varphi\big]_0^{2\pi} = -i\overline{\big(\psi(2\pi) - \psi(0)\big)}\,\varphi(0)

(using φ(2π)=φ(0)\varphi(2\pi) = \varphi(0)). For it to vanish for all such φ\varphi we need ψ(0)=ψ(2π)\psi(0) = \psi(2\pi), but nothing more, since φ(0)\varphi(0) is the only free quantity. So

DP∗∗={ψ∈H1([0,2π])∣ψ(0)=ψ(2π)}=DP∗.\mathcal{D}_{P^{**}} = \{\psi \in H^1([0, 2\pi]) \mid \psi(0) = \psi(2\pi)\} = \mathcal{D}_{P^*}.

and therefore P^j∗∗=P^j∗\hat{P}_j^{**} = \hat{P}_j^*, i.e. P^jcirc\hat{P}_j^{\text{circ}} is essentially self-adjoint.

Conclusion#

The momentum operator on the circle has a unique self-adjoint realization. Taking the closure of the naive operator on C1C^1 with periodic BC gives the operator −i∂x-i\partial_x with domain {ψ∈H1∣ψ(0)=ψ(2π)}\{\psi \in H^1 \mid \psi(0) = \psi(2\pi)\}.

Its eigenfunctions are einxe^{inx} with eigenvalues n∈Zn \in \mathbb{Z}, which is exactly the spectrum of momentum on S1S^1 one computes formally in any QM textbook. The point of all the machinery is to justify that calculation, by showing that the operator we are diagonalizing actually exists as a genuine observable.


7. The takeaway#

SettingSymmetric?Self-adjoint?Essentially self-adjoint?Self-adjoint extensions
Interval [0,2π][0,2\pi], ψ(0)=ψ(2π)=0\psi(0) = \psi(2\pi) = 0yesnonoone-parameter family θ∈U(1)\theta \in U(1)
Circle, ψ(0)=ψ(2π)\psi(0) = \psi(2\pi)yesno (regularity only)yesunique: the closure

The interval case is an example of something that comes up again and again in quantum mechanics. A perfectly innocent-looking differential operator can fail to define a unique observable, because the boundary conditions carry physical information that cannot be recovered from the formal expression alone. The circle is the happier case where everything works out, and it works out because the topology of the configuration space already fixes the boundary condition for us.

The general method for the harder cases is von Neumann’s. One counts the deficiency indices (n+,n−)(n_+, n_-), and gets a U(n+)U(n_+)-family of self-adjoint extensions when n+=n−n_+ = n_-, or no extensions at all when n+≠n−n_+ \neq n_-. This is the right tool for asking systematically whether a symmetric operator is a genuine observable, and if not, which data is missing.

We have also said nothing about why momentum is −i∂x-i\partial_x in the first place. That is the content of the Stone–von Neumann theorem, which recovers the Schrödinger representation up to unitary equivalence from the canonical commutation relation [x^,p^]=i[\hat x, \hat p] = i (plus an irreducibility assumption). The classical analogue is the Poisson bracket {x,p}=1\{x, p\} = 1. Quantization replaces Poisson brackets by commutators, and Stone–von Neumann says that, up to some mathematical niceness conditions, this prescription has essentially one realization. I will save that for some other day.


References#

  • B. C. Hall, Quantum Theory for Mathematicians, Springer (2013) — Chapters 7–10.
  • M. Reed and B. Simon, Methods of Modern Mathematical Physics, Volume I: Functional Analysis, Academic Press (1980) — especially Chapters VI–VIII.
  • F. Schuller, Lectures on Quantum Theory, University of Erlangen-Nürnberg, YouTube playlist (Lectures 6–9 cover this material).
  • P. Szekeres, A Course in Modern Mathematical Physics: Groups, Hilbert Space and Differential Geometry, Cambridge University Press (2004) — Chapters 13–14 for background reading.
Hermitian vs. Self-Adjoint: The Momentum Operator
https://rohankulkarni.me/posts/notes/case-study-momentum/
Author
Rohan Kulkarni
Published at
2021-12-01
License
CC BY-NC-SA 4.0
Share Post
Comments