Normalization

Formula

For a nonzero Hilbert-space vector $\lvert\chi\rangle$ with finite norm,

$$\lvert\psi\rangle = \frac{\lvert\chi\rangle} {\sqrt{\langle\chi\rvert\chi\rangle}}, \qquad \langle\psi\rvert\psi\rangle=1.$$

For a positive trace-class operator $\widetilde\rho$ with nonzero finite trace,

$$\rho = \frac{\widetilde\rho} {\operatorname{Tr}\widetilde\rho}, \qquad \operatorname{Tr}\rho=1.$$

These conditions fix total probability weight. They do not choose the global phase of a pure-state vector, make a non-normalizable function normalizable, or establish that a candidate lies in the domain of every observable.

At a glance

Object and convention	Normalization statement	Interpretation
Pure state	$\langle\psi\rvert\psi\rangle=1$	Physical state has total weight one
Density operator	$\rho\geq0$, $\operatorname{Tr}\rho=1$	General physical state
Orthonormal discrete basis	$\sum_n\lvert c_n\rvert^2=1$	Probabilities of complete basis outcomes sum to one
Coordinate wavefunction	$\int_X\lvert\psi(q)\rvert^2\,d\mu(q)=1$	Probability density is relative to $d\mu$
Nonorthogonal basis	$c^\dagger S c=1$	$S$ is the Gram or overlap matrix
Multi-component wavefunction	$\int_X\sum_\alpha\lvert\psi_\alpha(q)\rvert^2\,d\mu(q)=1$	Sum unresolved internal components
Continuum basis	$\langle\lambda\rvert\lambda'\rangle=\delta(\lambda-\lambda')$	Distributional basis convention, not a unit-norm state
Periodic box mode	$\langle n\rvert m\rangle=\delta_{nm}$	Finite-volume regulator
Conditional branch	$\operatorname{Tr}\widetilde\rho_a=p(a)\leq1$	Trace retains branch probability
Uniform numerical grid	$\sum_i\lvert\psi_i\rvert^2\Delta q=1$	Quadrature approximation to continuum norm

The word “normalization” is overloaded. Before using a formula, identify whether the object is a physical state, a generalized basis vector, a scattering solution, or a subnormalized conditional state.

Meaning

Normalization makes the Born weights of a complete measurement add to one. For projectors $\{P_a\}$ satisfying

$$\sum_aP_a=I,$$

one has

$$\sum_a \langle\psi\rvert P_a\lvert\psi\rangle = \langle\psi\rvert\psi\rangle.$$

Thus a unit vector produces a probability distribution. More generally, for a POVM $\{E_a\}$ and density operator $\rho$,

$$\sum_a \operatorname{Tr}(\rho E_a) = \operatorname{Tr}\rho.$$

Normalization removes an arbitrary positive scale. It does not remove global phase:

$$\lvert\psi\rangle \quad\hbox{and}\quad e^{i\alpha}\lvert\psi\rangle$$

are both normalized representatives of the same pure state.

Discrete representations

In an orthonormal basis,

$$\lvert\chi\rangle = \sum_n a_n\lvert n\rangle, \qquad \langle m\rvert n\rangle = \delta_{mn},$$

so

$$\lVert\chi\rVert^2 = \sum_n\lvert a_n\rvert^2.$$

The normalized coefficients are

$$c_n = \frac{a_n} {\sqrt{\sum_m\lvert a_m\rvert^2}}.$$

For an infinite expansion, the condition

$$\sum_n\lvert a_n\rvert^2<\infty$$

is essential. A divergent coefficient norm cannot be repaired by formal division.

Nonorthogonal bases

If $\{\lvert f_i\rangle\}$ is not orthonormal, define its Gram matrix

$$S_{ij} = \langle f_i\rvert f_j\rangle.$$

For

$$\lvert\chi\rangle = \sum_i a_i\lvert f_i\rangle,$$

the norm is

$$\lVert\chi\rVert^2 = a^\dagger S a.$$

Therefore

$$c = \frac{a}{\sqrt{a^\dagger S a}}, \qquad c^\dagger S c=1.$$

The Euclidean condition $c^\dagger c=1$ applies only when $S=I$, or after a consistent orthonormalizing transformation.

Coordinate wavefunctions

In a coordinate representation with measure $d\mu(q)$,

$$\psi(q) = \langle q\rvert\psi\rangle, \qquad \int_X \lvert\psi(q)\rvert^2 \,d\mu(q) = 1.$$

For one Cartesian coordinate,

$$\int_{-\infty}^{\infty} \lvert\psi(x)\rvert^2\,dx = 1.$$

For three Cartesian coordinates,

$$\int_{\mathbb R^3} \lvert\psi(\mathbf r)\rvert^2\,d^3r = 1.$$

In spherical coordinates,

$$\int_0^\infty dr \int_0^\pi d\theta \int_0^{2\pi}d\phi\, r^2\sin\theta \lvert\psi(r,\theta,\phi)\rvert^2 = 1.$$

The measure and density must be transformed together. Under a one-dimensional change of coordinates $y=f(x)$,

$$\lvert\psi_y(y)\rvert^2\,dy = \lvert\psi_x(x)\rvert^2\,dx,$$

so one possible phase convention gives

$$\psi_y(y) = \psi_x(x(y)) \sqrt{ \left\lvert \frac{dx}{dy} \right\rvert }.$$

The square root is the amplitude Jacobian. Applying only the classical density Jacobian to the amplitude gives the wrong norm.

Radial and multi-component forms

If

$$\psi(r,\theta,\phi) = R(r)Y_\ell^m(\theta,\phi)$$

and the spherical harmonic has unit angular norm, then

$$\int_0^\infty \lvert R(r)\rvert^2r^2\,dr = 1.$$

For the reduced radial function $u(r)=rR(r)$,

$$\int_0^\infty \lvert u(r)\rvert^2\,dr = 1.$$

Do not mix the measures for $R$ and $u$.

For a spinor or other multi-component wavefunction,

$$\int_X \sum_\alpha \lvert\psi_\alpha(q)\rvert^2 \,d\mu(q) = 1.$$

The components are not normally normalized one by one. Their total squared norm is one.

For an $N$-particle wavefunction in three dimensions,

$$\int \prod_{j=1}^N d^3r_j\, \lvert \Psi(\mathbf r_1,\ldots,\mathbf r_N) \rvert^2 = 1,$$

with sums over discrete spin labels when present. This is a density on configuration space.

Density operators and conditional states

A density operator representing a physical state satisfies

$$\rho\geq0, \qquad \rho^\dagger=\rho, \qquad \operatorname{Tr}\rho=1.$$

Trace one alone is not sufficient: a Hermitian trace-one matrix with a negative eigenvalue is not a valid state.

If a measurement outcome $a$ produces the unnormalized branch

$$\widetilde\rho_a = \mathcal I_a(\rho),$$

then

$$p(a) = \operatorname{Tr}\widetilde\rho_a.$$

When $p(a)>0$, the normalized conditional state is

$$\rho_a = \frac{\widetilde\rho_a}{p(a)}.$$

The subnormalized operator is useful precisely because its trace retains the outcome probability. Normalizing it too early can discard that information.

Continuum, box, and flux conventions

An ideal continuum eigenket is generally not a Hilbert-space vector. With a spectral label $\lambda$, one often uses

$$\langle\lambda\rvert\lambda'\rangle = \delta(\lambda-\lambda').$$

A physical wave packet is

$$\lvert\Psi\rangle = \int d\lambda\, c(\lambda) \lvert\lambda\rangle,$$

and its unit-norm condition is

$$\int d\lambda\, \lvert c(\lambda)\rvert^2 = 1.$$

Changing the continuum label changes the normalization. If $\lambda=\lambda(\eta)$ is one-to-one on the branch under consideration, then

$$\lvert\eta\rangle = \sqrt{ \left\lvert \frac{d\lambda}{d\eta} \right\rvert } \lvert\lambda(\eta)\rangle$$

obeys

$$\langle\eta\rvert\eta'\rangle = \delta(\eta-\eta').$$

Additional branch or channel labels must be kept when the relabeling is not one-to-one.

Three common continuum conventions answer different questions:

Convention	Defining statement	Main use
Delta normalization	$\langle\lambda\rvert\lambda'\rangle=\delta(\lambda-\lambda')$	Spectral expansions and wave packets
Box normalization	$\langle n\rvert m\rangle=\delta_{nm}$ in finite volume	Regulators and density-of-states limits
Flux normalization	Each channel carries prescribed incident or outgoing current	Scattering probabilities and $S$-matrices

Do not set $\delta(0)=1$, interpret a box mode as a literal infinite-volume state, or read a scattering probability from squared amplitudes without checking current factors.

Units and dimensions

Abstract normalized kets and density operators are dimensionless. Coordinate components can carry units because the probability measure must be dimensionless. If

$$\int_X \lvert\psi(q)\rvert^2 \,d\mu(q) = 1,$$

then

$$[\psi] = [d\mu]^{-1/2}.$$

Consequently,

$$[\psi(x)] = L^{-1/2}$$

in one spatial dimension and

$$[\psi(\mathbf r)] = L^{-3/2}$$

in three dimensions. For $u(r)=rR(r)$,

$$[R]=L^{-3/2}, \qquad [u]=L^{-1/2}.$$

Delta-normalized basis functions carry units determined by both their coordinate measure and spectral-label measure. A dimensional mismatch often reveals a missing Jacobian, grid spacing, or Fourier factor.

Numerical normalization

On a uniform grid $q_i$ with spacing $\Delta q$,

$$\int \lvert\psi(q)\rvert^2\,dq \approx \sum_i \lvert\psi_i\rvert^2\Delta q.$$

Normalize sampled wavefunction values by

$$\psi_i \longmapsto \frac{\psi_i} {\sqrt{ \sum_j \lvert\psi_j\rvert^2 \Delta q }}.$$

Many linear-algebra libraries instead return a vector $v$ with

$$\sum_i\lvert v_i\rvert^2=1.$$

For point samples on a uniform grid,

$$v_i \approx \sqrt{\Delta q}\,\psi(q_i).$$

On a nonuniform grid or with a higher-order quadrature rule, replace $\Delta q$ by positive weights $w_i$:

$$\sum_iw_i\lvert\psi_i\rvert^2=1.$$

For a finite nonorthogonal basis, use the overlap metric $c^\dagger S c=1$. The array norm is not automatically the physical norm.

Assumptions

• The object to be normalized is nonzero.
• An ordinary state vector has a finite Hilbert-space norm.
• A density-operator candidate is positive and has finite nonzero trace.
• The basis, integration measure, continuum label, or quadrature rule is specified.
• Discrete coefficient formulas use an orthonormal basis unless an overlap matrix is included.
• Coordinate formulas include all continuous coordinates and unresolved discrete components.
• Delta-normalized eigenkets are treated as distributions or generalized vectors, not ordinary physical states.
• Flux normalization is paired with the correct probability current for the Hamiltonian and channel.

Validity and limitations

Unit normalization is exact for physical states in standard quantum mechanics. It is not an approximation. The displayed repair formulas are valid only when their denominators are finite and strictly positive.

Normalization does not imply:

• finite energy, position, momentum, or variance;
• membership in the domain of an unbounded operator;
• satisfaction of the Hamiltonian’s boundary conditions;
• positivity of an arbitrary trace-one operator;
• localization of a wavefunction;
• purity of a density operator;
• completeness or orthogonality of a chosen basis;
• numerical convergence.

A normalized trial function can therefore be mathematically or physically inadmissible for the problem at hand.

Calculation checks

1. Positivity: every squared norm and density must be nonnegative.
2. Total weight: the complete sum, integral, or trace must equal one.
3. Measure: coordinate Jacobians and quadrature weights must be present.
4. Dimensions: the density times its measure must be dimensionless.
5. Representation independence: unitary basis changes must preserve the norm.
6. Continuum labels: relabeling requires the square root of the spectral Jacobian.
7. Density operators: eigenvalues must be nonnegative and sum to one.
8. Dynamics: unitary evolution preserves norm; unexpected drift in a closed numerical model signals an implementation or resolution problem.
9. Limiting procedures: box-size factors must cancel from physical continuum predictions.

Minimal worked uses

A finite-dimensional state

Let

$$\lvert\chi\rangle = \lvert0\rangle + (1+i)\lvert1\rangle.$$

In an orthonormal basis,

$$\lVert\chi\rVert^2 = 1+\lvert1+i\rvert^2 = 3.$$

One normalized representative is

$$\lvert\psi\rangle = \frac{1}{\sqrt3} \left[ \lvert0\rangle + (1+i)\lvert1\rangle \right].$$

Its computational-basis probabilities are $1/3$ and $2/3$.

An exponential bound-state shape

For

$$\chi(x) = A e^{-\kappa\lvert x\rvert}, \qquad \kappa>0,$$

normalization requires

$$\begin{aligned} 1 &= 2\lvert A\rvert^2 \int_0^\infty e^{-2\kappa x}\,dx\\ &= \frac{\lvert A\rvert^2}{\kappa}. \end{aligned}$$

Hence $\lvert A\rvert=\sqrt{\kappa}$. Normalization fixes the magnitude but not the phase of $A$.

Derivation and canonical home

Normalization is the canonical conceptual and mathematical treatment. It derives the probability sum rule, distinguishes rays from unit vectors, develops density-operator and conditional normalization, and states the infinite-dimensional caveats.

Normalization Conventions owns the detailed wave-mechanics comparison among bound-state, radial, box, delta, energy, flux, and numerical conventions. Plane Waves and Delta Normalization develops the continuum free-particle case.

Worked examples

• Normalization Examples
• Wavefunctions and Probability Density
• Gaussian Wave Packets
• Matrix Diagonalization

Common mistakes

• Dividing the zero vector or an infinite-norm expression by a formal norm.
• Treating global phase as fixed by normalization.
• Using $\sum_i\lvert c_i\rvert^2=1$ in a nonorthogonal basis.
• Forgetting coordinate Jacobians such as $r^2\sin\theta$.
• Normalizing every spinor component separately.
• Confusing $R(r)$ with the reduced radial function $u(r)=rR(r)$.
• Treating $\lvert\psi(x)\rvert^2$ as a point probability instead of a density.
• Applying square normalization directly to an infinite-volume plane wave.
• Mixing $p$- and $k$-normalized continuum states.
• Normalizing a conditional branch before recording its probability.
• Assuming trace one makes a matrix a valid density operator.
• Comparing an unweighted array norm with a continuum or nonorthogonal-basis norm.
• Renormalizing at every numerical time step to hide a nonunitary integrator.
• Assuming a normalized state has finite expectation values for all observables.

Related formulas

• Born Rule
• Expectation Value
• Variance
• Probability Current
• Density-Matrix Expectation
• Partial Trace
• Generalized Eigenvectors
• L² Spaces

References

• R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer, 1994, chs. 1 and 4.
• J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020, ch. 1.
• D. J. Griffiths and D. F. Schroeter, Introduction to Quantum Mechanics, 3rd ed., Cambridge University Press, 2018, chs. 1 and 2.
• M. Reed and B. Simon, Methods of Modern Mathematical Physics I: Functional Analysis, rev. ed., Academic Press, 1980, chs. II and VII.
• J. R. Taylor, Scattering Theory: The Quantum Theory of Nonrelativistic Collisions, Dover, 2006, chs. 1 and 3.

Exercises

Exercise 1: normalize complex amplitudes

Normalize

$$\lvert\chi\rangle = 2\lvert0\rangle - i\lvert1\rangle + \lvert2\rangle$$

in an orthonormal basis, and give the three basis probabilities.

Solution

The squared norm is

$$\lVert\chi\rVert^2 = \lvert2\rvert^2 + \lvert-i\rvert^2 + \lvert1\rvert^2 = 6.$$

Thus

$$\lvert\psi\rangle = \frac{1}{\sqrt6} \left( 2\lvert0\rangle - i\lvert1\rangle + \lvert2\rangle \right).$$

The probabilities are

$$p_0=\frac{2}{3}, \qquad p_1=\frac{1}{6}, \qquad p_2=\frac{1}{6}.$$

They sum to one.

Exercise 2: radial measure

A spherically symmetric wavefunction has the form

$$\psi(\mathbf r) = A e^{-r/a}, \qquad a>0.$$

Find the magnitude of $A$.

Solution

Spherical symmetry gives

$$1 = 4\pi\lvert A\rvert^2 \int_0^\infty r^2e^{-2r/a}\,dr.$$

Using

$$\int_0^\infty r^2e^{-\beta r}\,dr = \frac{2}{\beta^3}$$

with $\beta=2/a$,

$$\int_0^\infty r^2e^{-2r/a}\,dr = \frac{a^3}{4}.$$

Therefore

$$1 = \pi a^3\lvert A\rvert^2, \qquad \lvert A\rvert = \frac{1}{\sqrt{\pi a^3}}.$$

The overall phase remains arbitrary.

Exercise 3: a nonorthogonal basis

Two normalized basis vectors satisfy

$$\langle f_1\rvert f_2\rangle = \langle f_2\rvert f_1\rangle = s, \qquad 0<s<1.$$

Normalize $\lvert\chi\rangle=\lvert f_1\rangle+\lvert f_2\rangle$.

Solution

The overlap matrix is

$$S = \begin{pmatrix} 1&s\\ s&1 \end{pmatrix}.$$

For $a=(1,1)^{\mathsf T}$,

$$a^\dagger S a = 2+2s.$$

Hence

$$\lvert\psi\rangle = \frac{ \lvert f_1\rangle+\lvert f_2\rangle }{ \sqrt{2(1+s)} }.$$

Using $1/\sqrt2$ would be correct only for $s=0$.