Relation to Berry Geometry

Projective Hilbert space explains why a pure state is a ray, while Berry geometry explains how the phase of a chosen vector twists as a ray varies over a parameter space. The two subjects are not merely analogous. Berry data are obtained by pulling the universal phase-bundle geometry of projective Hilbert space back along a family of quantum states.

The key map is

$$\iota_n:\mathcal M\to\mathbb P(\mathcal H), \qquad R\longmapsto[n(R)],$$

where $\mathcal M$ is a parameter manifold and $\lvert n(R)\rangle$ is a normalized nondegenerate eigenstate chosen locally. The Berry connection and curvature live over $\mathcal M$, but they originate from the phase bundle over $\mathbb P(\mathcal H)$.

This page establishes that bridge. The physical adiabatic effect remains canonical in Berry Phase, and the detailed gauge and curvature formulas remain canonical in Berry Connection and Berry Curvature.

Two Spaces That Must Be Distinguished

Berry discussions involve two manifolds that are easy to conflate.

The first is projective Hilbert space:

$$\mathbb P(\mathcal H) = S(\mathcal H)/U(1),$$

the space of all pure-state rays. A point is a physical state $[\psi]$.

The second is a parameter space $\mathcal M$, with coordinates

$$R=(R^1,\ldots,R^d),$$

on which a Hamiltonian depends:

$$H=H(R).$$

An isolated eigenstate family gives a ray-valued map

$$R\longmapsto[n(R)].$$

Parameter space need not resemble projective Hilbert space. It may be a sphere of field directions, a Brillouin zone, a space of molecular coordinates, or another control manifold. The eigenstate map traces a submanifold or immersed image inside the much larger pure-state space.

This distinction resolves several common puzzles:

• the Fubini–Study metric is intrinsic to ray space;
• the quantum metric on parameter space is its pullback;
• the universal phase connection lives over ray space;
• the Berry connection is its pullback in a chosen eigenvector gauge;
• degeneracies can obstruct a smooth eigenstate map even when the ambient projective space is regular.

The Universal Phase Bundle

Normalized vectors form the unit sphere

$$S(\mathcal H) = \{ \lvert\psi\rangle: \langle\psi\vert\psi\rangle=1 \}.$$

The projection

$$U(1) \longrightarrow S(\mathcal H) \longrightarrow \mathbb P(\mathcal H)$$

forgets global phase. Each ray has a circle of normalized representatives

$$\lvert\psi\rangle \longmapsto e^{i\chi}\lvert\psi\rangle.$$

A real-valued connection one-form in the convention used by the Berry pages is

$$\mathcal A = i\langle\psi\vert d\psi\rangle.$$

Normalization implies that $\langle\psi\vert d\psi\rangle$ is purely imaginary, so $\mathcal A$ is real. Under a phase change,

$$\lvert\psi\rangle \longmapsto e^{i\chi}\lvert\psi\rangle,$$

the connection transforms as

$$\mathcal A \longmapsto \mathcal A-d\chi.$$

The connection is therefore gauge dependent, as any local rule for comparing phases must be. Its kernel defines horizontal directions:

$$\mathcal A(\dot\psi)=0 \quad\Longleftrightarrow\quad \langle\psi\vert\dot\psi\rangle=0.$$

A horizontal lift is the vector representative that accumulates no local phase according to this connection.

Parallel Transport and Holonomy

Let $[\psi(t)]$ be a path in projective Hilbert space. Any normalized lift $\lvert\psi(t)\rangle$ may be rephased:

$$\lvert\psi(t)\rangle \longmapsto e^{i\chi(t)} \lvert\psi(t)\rangle.$$

Along an open interval, one can choose $\chi(t)$ so that

$$\langle\psi\vert\dot\psi\rangle=0.$$

This is the parallel-transport gauge. It removes phase motion tangent to the $U(1)$ fiber while leaving the ray path unchanged.

If the ray path closes,

$$[\psi(T)]=[\psi(0)],$$

a horizontal lift need not close as a vector. Instead,

$$\lvert\psi_{\rm hor}(T)\rangle = e^{i\gamma_{\rm geom}} \lvert\psi_{\rm hor}(0)\rangle.$$

The phase factor is the holonomy of the connection around the loop. It is global information about the path, despite the horizontal condition being local.

The detailed construction belongs to Parallel Transport and Holonomy. The role of this page is to identify their base geometry as the phase bundle over ray space.

Horizontal Schrödinger Motion

For a normalized Schrödinger solution,

$$\lvert\dot\psi\rangle = -\frac{i}{\hbar}H\lvert\psi\rangle,$$

the connection evaluated on the velocity is

$$\mathcal A(\dot\psi) = i\langle\psi\vert\dot\psi\rangle = \frac{\langle H\rangle_\psi}{\hbar}.$$

Thus the full Schrödinger lift is generally not horizontal. Its vertical part is the familiar dynamical phase rotation. Removing that part gives

$$\lvert\dot\psi\rangle_{\rm hor} = -\frac{i}{\hbar} \left( H-\langle H\rangle_\psi \right) \lvert\psi\rangle,$$

which is exactly the Hamiltonian vector field derived in Hamiltonian Flow on Projective Hilbert Space.

Suppose a Schrödinger evolution is cyclic in ray space:

$$\lvert\psi(T)\rangle = e^{i\alpha} \lvert\psi(0)\rangle.$$

The geometric phase of the ray loop is

$$\gamma_{\rm geom} = \alpha + \frac{1}{\hbar} \int_0^T dt\, \langle H\rangle_{\psi(t)} \pmod{2\pi}.$$

The second term removes the dynamical phase

$$\gamma_{\rm dyn} = -\frac{1}{\hbar} \int_0^T dt\, \langle H\rangle_{\psi(t)}.$$

This projective geometric phase does not require adiabatic evolution. In the adiabatic eigenstate setting it reduces to Berry’s phase. The more general cyclic construction is often called the Aharonov–Anandan phase.

An energy eigenstate provides a useful limiting check. Its ray is stationary,

$$\alpha=-\frac{ET}{\hbar}, \qquad \langle H\rangle=E,$$

so

$$\gamma_{\rm geom}=0.$$

The vector acquires dynamical phase, but the ray traces no loop with enclosed geometry.

Berry Connection as a Pullback

Let $\mathcal M$ be a region of parameter space on which a normalized eigenvector $\lvert n(R)\rangle$ can be chosen smoothly. This choice is a local section of the pulled-back phase bundle.

Pulling the universal connection back along the eigenstate map gives

$$A_n = \iota_n^*\mathcal A = i\langle n(R)\vert d n(R)\rangle.$$

In coordinates,

$$A_i^{(n)} = i\langle n\vert\partial_i n\rangle.$$

This is precisely the Berry connection. A different local eigenvector phase,

$$\lvert n(R)\rangle \longmapsto e^{i\chi(R)} \lvert n(R)\rangle,$$

changes the local one-form by

$$A_n\longmapsto A_n-d\chi.$$

The connection is not an extra structure attached by hand to parameter space. It is inherited from the universal rule for comparing phases of neighboring rays.

The corresponding eigenstate line bundle is the pullback of the tautological line bundle over projective Hilbert space:

$$\mathcal L_n = \iota_n^*\mathcal L_{\rm taut}.$$

This statement is the global version of the local formula for $A_n$. It explains why no globally smooth eigenvector need exist even when local gauges do.

Berry Curvature and the Fubini–Study Form

The curvature of the universal connection is

$$\mathcal F=d\mathcal A.$$

For horizontal tangent vectors $u$ and $v$,

$$\mathcal F(u,v) = -2\operatorname{Im} \langle u\vert v\rangle.$$

The Hamiltonian-normalized Fubini–Study symplectic form used in this chapter is

$$\omega_{\rm FS}(u,v) = 2\hbar\, \operatorname{Im} \langle u\vert v\rangle.$$

Therefore, with the connection convention $\mathcal A=i\langle\psi\vert d\psi\rangle$,

$$\mathcal F = -\frac{1}{\hbar}\omega_{\rm FS}.$$

Changing the sign convention for the connection changes this displayed sign, not the underlying geometry.

Pulling the curvature back to parameter space gives

$$F_n = dA_n = \iota_n^*\mathcal F = -\frac{1}{\hbar} \iota_n^*\omega_{\rm FS}.$$

This is the direct relation between Berry curvature and the symplectic geometry of projective Hilbert space.

The metric and curvature can be packaged in the quantum geometric tensor

$$Q_{ij}^{(n)} = \langle\partial_i n\vert \left( I-P_n \right) \vert\partial_j n\rangle, \qquad P_n=\lvert n\rangle\langle n\rvert.$$

Its real and imaginary parts are

$$g_{ij}^{(n)} = \operatorname{Re}Q_{ij}^{(n)}, \qquad F_{ij}^{(n)} = -2\operatorname{Im}Q_{ij}^{(n)}.$$

Thus the quantum metric is the pullback of the Fubini–Study metric, while Berry curvature is the pullback of its Kähler two-form up to the stated normalization and sign.

From Projective Geometry to Berry Phase

Assume $H(R)$ has an isolated nondegenerate eigenvalue $E_n(R)$ along a closed parameter loop $C$. Under the hypotheses of the adiabatic theorem, a state prepared in the corresponding eigenspace follows the ray

$$R(t) \longmapsto [n(R(t))].$$

The geometric holonomy of this ray loop is

$$\gamma_n[C] = \oint_C A_n \pmod{2\pi}.$$

Equivalently, when $C$ bounds a surface $\Sigma$ on which the relevant data are regular,

$$\gamma_n[C] = \int_\Sigma F_n \pmod{2\pi}.$$

These formulas are stated here only to complete the bridge. Their derivation, gauge qualifications, adiabatic assumptions, spin example, and applications belong to the canonical pages in Symmetry, Angular Momentum, and Spin.

The logical chain is:

$$\begin{gathered} \text{phase bundle over }\mathbb P(\mathcal H) \\ \downarrow\ \text{pull back along }\iota_n \\ \text{eigenstate line bundle over }\mathcal M \\ \downarrow\ \text{connection and curvature} \\ \text{Berry holonomy around }C. \end{gathered}$$

Canonical Homes

Use the following pages for the detailed subjects:

• Berry Phase owns the cyclic adiabatic phenomenon and physical interpretation.
• Dynamical vs Geometric Phase owns the phase decomposition in the adiabatic setting.
• Berry Connection owns the local one-form, gauge transformation, and line integral.
• Berry Curvature owns the gauge-invariant two-form, projector formula, nearby-level formula, and applications.
• Parallel Transport owns the horizontal-gauge construction along paths.
• Holonomy owns closed-loop mismatch and its abelian and nonabelian extensions.
• U(1) Bundles and Quantum Phase owns the bundle language.
• Berry Connection as a Mathematical Object owns the formal connection, curvature, and projector derivations.

Common Mistakes

• Treating parameter space as identical to projective Hilbert space rather than mapping it into ray space.
• Calling a chosen eigenvector a globally defined physical state without checking gauge patches.
• Confusing the gauge-dependent connection with the gauge-invariant curvature.
• Forgetting that the connection sign convention controls the displayed sign relating curvature to $\omega_{\rm FS}$.
• Assuming every geometric phase requires adiabatic evolution. Adiabatic Berry phase is a special case of cyclic projective holonomy.
• Calling the dynamical phase geometric merely because it is written using the phase connection.
• Ignoring degeneracies, where an isolated eigenline may cease to exist.
• Applying the abelian formulas unchanged to a degenerate eigenspace.
• Re-deriving Berry curvature in projective language without respecting its canonical home.

References

• M. V. Berry, “Quantal phase factors accompanying adiabatic changes,” Proceedings of the Royal Society A 392, 45–57, 1984.
• B. Simon, “Holonomy, the quantum adiabatic theorem, and Berry’s phase,” Physical Review Letters 51, 2167–2170, 1983.
• Y. Aharonov and J. Anandan, “Phase change during a cyclic quantum evolution,” Physical Review Letters 58, 1593–1596, 1987.
• J. P. Provost and G. Vallee, “Riemannian structure on manifolds of quantum states,” Communications in Mathematical Physics 76, 289–301, 1980.
• A. Ashtekar and T. A. Schilling, “Geometrical formulation of quantum mechanics,” in On Einstein’s Path, Springer, 1999; arXiv:gr-qc/9706069.
• I. Bengtsson and K. Życzkowski, Geometry of Quantum States, 2nd ed., Cambridge University Press, 2017.

Exercises

1. Derive the gauge transformation of the universal phase connection.

Solution

Let

$$\lvert\psi'\rangle = e^{i\chi}\lvert\psi\rangle.$$

Then

$$d\lvert\psi'\rangle = e^{i\chi} \left( i\,d\chi\lvert\psi\rangle + d\lvert\psi\rangle \right).$$

Therefore

$$\begin{aligned} \mathcal A' &= i\langle\psi'\vert d\psi'\rangle\\ &= i \left( i\,d\chi + \langle\psi\vert d\psi\rangle \right)\\ &= \mathcal A-d\chi. \end{aligned}$$

The horizontal subspace changes covariantly with the local phase convention.

2. Show that the curvature of the universal connection is proportional to the Fubini–Study symplectic form.

Solution

For

$$\mathcal A = i\langle\psi\vert d\psi\rangle,$$

the curvature evaluated on horizontal vectors $u,v$ is

$$\begin{aligned} \mathcal F(u,v) &= i \left( \langle u\vert v\rangle - \langle v\vert u\rangle \right)\\ &= -2\operatorname{Im} \langle u\vert v\rangle. \end{aligned}$$

Because

$$\omega_{\rm FS}(u,v) = 2\hbar\, \operatorname{Im} \langle u\vert v\rangle,$$

one obtains

$$\mathcal F = -\frac{1}{\hbar}\omega_{\rm FS}.$$

3. Explain why the projector in the quantum geometric tensor makes it gauge invariant.

Solution

Under

$$\lvert n\rangle \longmapsto e^{i\chi}\lvert n\rangle,$$

the derivative gains a vertical term:

$$\lvert\partial_i n\rangle \longmapsto e^{i\chi} \left( \lvert\partial_i n\rangle + i(\partial_i\chi)\lvert n\rangle \right).$$

The operator

$$I-P_n$$

annihilates $\lvert n\rangle$, so it removes this gauge-dependent vertical term. The remaining horizontal derivatives acquire only a common phase, which cancels between bra and ket in

$$Q_{ij}^{(n)} = \langle\partial_i n\vert (I-P_n) \vert\partial_j n\rangle.$$

Thus both its real metric part and imaginary curvature part are gauge invariant.

4. Derive the geometric phase of a cyclic Schrödinger evolution.

Solution

Suppose

$$\lvert\psi(T)\rangle = e^{i\alpha} \lvert\psi(0)\rangle.$$

Define a rephased lift

$$\lvert\widetilde\psi(t)\rangle = e^{i\chi(t)} \lvert\psi(t)\rangle$$

with $\chi(0)=0$. Its connection along the path is

$$\widetilde{\mathcal A}_t = \mathcal A_t-\dot\chi.$$

For Schrödinger evolution,

$$\mathcal A_t = \frac{\langle H\rangle_{\psi(t)}}{\hbar}.$$

Choosing

$$\dot\chi = \frac{\langle H\rangle_{\psi(t)}}{\hbar}$$

makes the lift horizontal. At the endpoint,

$$\lvert\widetilde\psi(T)\rangle = \exp \left[ i\alpha + \frac{i}{\hbar} \int_0^Tdt\, \langle H\rangle_{\psi(t)} \right] \lvert\psi(0)\rangle.$$

The exponent is the geometric holonomy:

$$\gamma_{\rm geom} = \alpha + \frac{1}{\hbar} \int_0^Tdt\, \langle H\rangle_{\psi(t)} \pmod{2\pi}.$$

5. Why can the pulled-back quantum metric be degenerate even though the Fubini–Study metric is nondegenerate on projective Hilbert space?

Solution

The pullback measures only how the ray changes under parameter variations. If a nonzero parameter tangent vector $v\in T_R\mathcal M$ satisfies

$$d\iota_n(v)=0,$$

then moving in that parameter direction does not change the eigenray. Its pulled-back length is

$$g^{(n)}(v,v) = g_{\rm FS} \left( d\iota_n(v), d\iota_n(v) \right) =0.$$

The ambient Fubini–Study metric remains nondegenerate; the degeneracy reflects redundancy or loss of rank in the parameter-to-state map.