Normalization Conventions

Normalization turns probability amplitudes into correctly scaled states. In wave mechanics there is no single convention that covers every use case. Bound states, plane waves, scattering states, radial wavefunctions, periodic boxes, and numerical grid vectors are normalized differently because they represent different mathematical objects.

The rule is:

State the normalization convention before interpreting amplitudes as probabilities or currents.

Required background. Wavefunctions and Probability Density supplies the coordinate measure and Born-density interpretation.

Helpful background. Born Rule for Continuous Outcomes explains continuous probability measures. Probability Current supplies the channel flux used in scattering normalization.

A Practical Decision Process

Before calculating a normalization constant, identify four pieces of structure:

1. The state type. A bound wavefunction is an element of the Hilbert space; an exact plane wave is a generalized eigenfunction; a wave packet is again a normalizable state.
2. The integration measure. Cartesian, spherical, reduced-radial, and numerical-grid representations use different measures.
3. The spectral label. Continuum states labeled by $p$, $k$, or $E$ carry different Dirac-delta normalizations.
4. The intended observable. State normalization, energy normalization, and unit-flux normalization answer different questions.

Only then choose a constant. A compact model-by-model normalization table is a later lookup projection; the present page owns the conventions and their conversion rules.

Bound-State Normalization

A normalizable bound state in one dimension satisfies

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

or, on an interval $[a,b]$,

$$\int_a^b\lvert\psi(x)\rvert^2\,dx=1.$$

This is square normalization. It applies to ordinary physical states with total probability one. For a discrete orthonormal set of bound states,

$$\int \psi_m^*(x)\psi_n(x)\,dx=\delta_{mn}.$$

The Kronecker delta appears because the labels $m,n$ are discrete.

Normalization and orthogonality are distinct. Multiplying each eigenfunction by a constant can give unit norm, but degenerate eigenfunctions may still need to be orthogonalized within their degenerate eigenspace. Once an orthonormal discrete basis is chosen,

$$\psi(x)=\sum_n c_n\psi_n(x), \qquad \sum_n\lvert c_n\rvert^2=1.$$

For example, the one-dimensional bound-state ansatz

$$\psi(x)=A e^{-\kappa\lvert x\rvert}, \qquad \kappa\gt0,$$

has

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

Thus $\lvert A\rvert=\sqrt{\kappa}$. Normalization fixes the magnitude of $A$, while its overall phase remains arbitrary.

Three-Dimensional Normalization

In three dimensions,

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

In spherical coordinates this is

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

For separated central-potential states,

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

one usually normalizes angular and radial parts by

$$\int \lvert Y_\ell^m\rvert^2\,d\Omega=1, \qquad \int_0^\infty \lvert R(r)\rvert^2r^2\,dr=1.$$

If $u(r)=rR(r)$ is used, then radial normalization becomes

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

When the angular factor is normalized, the probability of finding the particle in a spherical shell is

$$dP = \lvert R(r)\rvert^2r^2\,dr = \lvert u(r)\rvert^2\,dr.$$

Thus $\lvert R(r)\rvert^2$ is not by itself a radial probability density with respect to $dr$; $\lvert u(r)\rvert^2$ is. Confusing $R(r)$ and $u(r)$ is one of the most common normalization errors in central-potential problems.

If angular, spin, or channel labels are present, sum over every discrete component as well as integrating over position. For a two-component spinor in three dimensions, for example,

$$\int d^3r\, \left( \lvert\psi_1(\mathbf r)\rvert^2 +\lvert\psi_2(\mathbf r)\rvert^2 \right)=1.$$

Box Normalization

Plane waves on the full line are not square-normalizable. One practical workaround is to place the system in a large box of length $L$ and impose periodic boundary conditions. In one dimension,

$$\psi_n(x)=\frac{1}{\sqrt L}e^{ik_nx}, \qquad k_n=\frac{2\pi n}{L}.$$

Then

$$\int_0^L \lvert\psi_n(x)\rvert^2\,dx=1.$$

Box normalization is useful for counting states, deriving density of states factors, and regulating continuum calculations. Physical continuum results should not depend on the artificial box after the limit $L\to\infty$ is taken correctly.

For a rectangular three-dimensional periodic box with volume $\mathcal V=L_xL_yL_z$,

$$\psi_{\mathbf n}(\mathbf r) =\frac{1}{\sqrt{\mathcal V}} e^{i\mathbf k_{\mathbf n}\cdot\mathbf r}, \qquad k_i=\frac{2\pi n_i}{L_i}.$$

The continuum limit converts a sum over allowed wavevectors into an integral:

$$\sum_{\mathbf k} \longrightarrow \frac{\mathcal V}{(2\pi)^3} \int d^3k.$$

This replacement is valid only when the level spacing is small on the scale over which the rest of the summand changes. Boundary corrections and discrete low-energy levels can matter in finite systems.

Delta Normalization

Continuum eigenstates are often normalized to Dirac delta functions. For momentum eigenstates in one dimension, a common convention is

$$\langle p\vert p'\rangle=\delta(p-p').$$

In position representation,

$$\langle x\vert p\rangle =\frac{1}{\sqrt{2\pi\hbar}}e^{ipx/\hbar},$$

which gives

$$\int_{-\infty}^{\infty} \langle p\vert x\rangle\langle x\vert p'\rangle\,dx =\delta(p-p').$$

If states are labeled by wavenumber $k$ instead of momentum $p=\hbar k$, the normalization changes. Since

$$\delta(p-p')=\frac{1}{\hbar}\delta(k-k'),$$

one must track which continuum label is being used.

The corresponding $k$-normalized convention is

$$\langle k|k'\rangle=\delta(k-k'), \qquad \langle x|k\rangle =\frac{1}{\sqrt{2\pi}}e^{ikx},$$

with

$$|k\rangle=\sqrt{\hbar}\,|p=\hbar k\rangle.$$

More generally, if a monotonic relabeling $\lambda=\lambda(\eta)$ starts from $\langle\lambda|\lambda'\rangle=\delta(\lambda-\lambda')$, then

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

is delta-normalized in $\eta$. When an energy has multiple branches, such as right- and left-moving free waves, the branch or channel label must also be retained.

An exact continuum eigenstate is not an ordinary unit-norm state. A physical wave packet has the expansion

$$|\Psi\rangle =\int dp\,c(p)|p\rangle, \qquad \int dp\,\lvert c(p)\rvert^2=1.$$

The Dirac delta is what makes the packet norm reduce to the integral of $\lvert c(p)\rvert^2$. Expressions such as $\delta(0)$ signal that an ideal generalized eigenstate has been treated as though it were square-normalizable.

Connecting Box and Delta Normalization

For a periodic box, adjacent wave numbers are separated by

$$\Delta k=\frac{2\pi}{L}.$$

Let $|n\rangle_{\mathrm{box}}$ have unit norm and let $|k\rangle_\delta$ obey $\langle k|k'\rangle=\delta(k-k')$. At the sampled values $k_n$, the conventions are related by

$$|k_n\rangle_\delta \approx \frac{1}{\sqrt{\Delta k}} |n\rangle_{\mathrm{box}}.$$

Indeed, the position-space amplitude becomes

$$\frac{1}{\sqrt{\Delta k}} \frac{1}{\sqrt L} =\frac{1}{\sqrt{2\pi}},$$

which is the $k$-normalized plane-wave convention. For a wave packet, the continuum coefficient $c(k)$ and box coefficients $c_n$ satisfy

$$c_n\approx\sqrt{\Delta k}\,c(k_n),$$

so that

$$\sum_n\lvert c_n\rvert^2 \longrightarrow \int dk\,\lvert c(k)\rvert^2.$$

This relation is the reliable way to move factors of $L$ and $2\pi$ through a continuum limit; guessing them from dimensional appearance is error-prone.

Flux Normalization

In scattering, reflection and transmission probabilities are ratios of probability currents, not just squared amplitude ratios. A plane wave

$$\psi(x)=Ae^{ikx}$$

has current

$$j=\frac{\hbar k}{m}\lvert A\rvert^2$$

for $k>0$. If incident and transmitted waves have different wavenumbers, then transmission includes a velocity factor:

$$T=\frac{j_{\text{trans}}}{j_{\text{inc}}} =\frac{v_{\text{trans}}}{v_{\text{inc}}} \lvert t\rvert^2,$$

where the last equality assumes incident amplitude one and the usual plane-wave convention in each channel. The velocity is the group velocity

$$v=\frac{1}{\hbar}\frac{dE}{dk}.$$

Flux normalization chooses a spatial amplitude proportional to $1/\sqrt{\lvert v\rvert}$, so that each incoming or outgoing channel carries unit-magnitude current. In that basis, squared scattering-matrix elements can be interpreted directly as channel probabilities.

Energy normalization is closely related but not identical. Starting from $k$-normalized states on a single monotonic branch,

$$|E,\sigma\rangle = \frac{1}{\sqrt{\hbar\lvert v\rvert}} |k,\sigma\rangle,$$

where $\sigma$ distinguishes propagation directions or channels. These states obey

$$\langle E,\sigma|E',\sigma'\rangle = \delta_{\sigma\sigma'}\delta(E-E').$$

With $\langle x|k\rangle=(2\pi)^{-1/2}e^{ikx}$, an energy-normalized traveling wave carries current magnitude $1/(2\pi\hbar)$ rather than one. Unit-flux and unit-energy conventions therefore differ by the fixed factor $\sqrt{2\pi\hbar}$ in this Fourier convention. Always read the stated inner product before interpreting an amplitude.

The change from $k$ to $E$ is branchwise. At a threshold where $v=(1/\hbar)dE/dk$ vanishes, the Jacobian is singular and this traveling-wave normalization must be interpreted through an appropriate limiting spectral measure rather than by dividing naively by $\sqrt{|v|}$.

Numerical Grid Normalization

Numerical wavefunctions are often represented by values on grid points $x_i$ with spacing $\Delta x$. The continuum norm is approximated by

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

Thus the normalized grid values should satisfy

$$\sum_i \lvert\psi_i\rvert^2\Delta x=1.$$

If a numerical library normalizes vectors using $\sum_i \lvert v_i\rvert^2=1$, then on a uniform grid

$$v_i\approx\sqrt{\Delta x}\,\psi(x_i), \qquad \psi(x_i)\approx\frac{v_i}{\sqrt{\Delta x}}.$$

This distinction matters when comparing numerical eigenvectors with analytic wavefunctions or computing a pointwise probability density.

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

$$\sum_i w_i\lvert\psi_i\rvert^2=1.$$

In a nonorthogonal basis $\{f_i\}$, the overlap matrix

$$S_{ij}=\langle f_i|f_j\rangle$$

defines the norm. If $\psi=\sum_i c_i f_i$, then

$$c^\dagger S c=1,$$

and the stationary problem generally has the form $Hc=ESc$. Euclidean normalization $c^\dagger c=1$ is correct only when the basis is orthonormal or the overlap has already been absorbed into transformed coefficients.

For time propagation, monitor the physically weighted norm, not merely the array norm. A stable calculation should preserve it up to the expected integration error unless the Hamiltonian deliberately contains absorbing boundaries or another nonunitary element.

Units Of Wavefunctions

The units of a square-normalized wavefunction follow from the measure. If $q$ denotes coordinates and $d\mu(q)$ has units $U$, then

$$\int d\mu(q)\,\lvert\psi(q)\rvert^2=1 \quad\Longrightarrow\quad [\psi]=U^{-1/2}.$$

In common position representations:

• in one dimension, $\psi$ has units length$^{-1/2}$;
• in three dimensions, $\psi$ has units length$^{-3/2}$;
• a radial function $R(r)$ has units length$^{-3/2}$;
• the reduced radial function $u(r)=rR(r)$ has units length$^{-1/2}$.

For $N$ distinguishable particles in three spatial dimensions, a configuration-space wavefunction has units length$^{-3N/2}$. Its modulus squared is a density on configuration space, not generally an ordinary density in three-dimensional physical space.

The units of delta-normalized generalized eigenfunctions also depend on the spectral label. For example, $\langle x|p\rangle$ and $\langle x|k\rangle$ cannot have the same dimensions because $\delta(p-p')$ and $\delta(k-k')$ do not. Dimensional analysis is an excellent audit, but it cannot choose among valid Fourier conventions by itself.

Normalization Is Not Localization

A normalized state can be arbitrarily broad, and a sharply localized ideal such as $\delta(x-x_0)$ is not a square-normalizable wavefunction. Likewise, rescaling a non-normalizable plane wave cannot turn it into a Hilbert-space vector on the full line. The proper physical replacements are normalized wave packets or controlled box limits.

Normalization also does not establish that a candidate lies in the domain of the Hamiltonian. Boundary conditions, differentiability, and finite expectation values may impose additional requirements. Boundary Conditions develops those admissibility checks.

Common Mistakes

• Normalizing a plane wave on the real line as if $\int \lvert e^{ikx}\rvert^2 dx$ were finite.
• Interpreting the constant modulus of a delta-normalized plane wave as an ordinary position probability density.
• Mixing $k$-normalization and $p$-normalization without the factor of $\hbar$.
• Relabeling continuum states by energy without the Jacobian or a propagation-direction label.
• Forgetting the $r^2$ measure in radial normalization.
• Normalizing $R(r)$ with $dr$ after already defining $u(r)=rR(r)$.
• Comparing scattering amplitudes without current or velocity factors.
• Treating energy normalization and unit-flux normalization as identical conventions.
• Treating a box-normalized plane wave as if the artificial box were physical.
• Forgetting $\Delta x$ factors when normalizing numerical grid wavefunctions.
• Using $c^\dagger c=1$ in a nonorthogonal basis instead of $c^\dagger S c=1$.
• Normalizing each component of a spinor separately rather than normalizing their summed probability.

Where this is used

• Wavefunctions and Probability Density explains density and probability.
• Probability Current uses velocity and flux normalization in scattering channels.
• Boundary Conditions distinguishes normalizability from operator-domain admissibility.
• Time-Independent Schrödinger Equation combines discrete and continuum spectral sectors.

Exercises

1. Normalize the periodic plane wave $\psi_n(x)=Ae^{i2\pi nx/L}$ on $0\le x<L$. Then use $\Delta k=2\pi/L$ to recover the position wavefunction of a $k$-normalized continuum state.

Solution

The normalization condition is

$$1=\int_0^L \lvert A\rvert^2\,dx=\lvert A\rvert^2L.$$

Thus $\lvert A\rvert=1/\sqrt L$. A standard phase choice is $A=1/\sqrt L$. Since

$$|k_n\rangle_\delta \approx \frac{|n\rangle_{\mathrm{box}}}{\sqrt{\Delta k}},$$

its position wavefunction is

$$\langle x|k_n\rangle_\delta \approx \frac{e^{ik_nx}}{\sqrt{L\Delta k}} =\frac{e^{ik_nx}}{\sqrt{2\pi}}.$$

2. Let a wave packet be written as

$$|\Psi\rangle =\int dp\,c_p(p)|p\rangle =\int dk\,c_k(k)|k\rangle,$$

where $p=\hbar k$, $\langle p|p'\rangle=\delta(p-p')$, and $\langle k|k'\rangle=\delta(k-k')$. Find $c_k(k)$ in terms of $c_p(p)$ and verify that the two normalization integrals agree.

Solution

The normalized basis states obey $|k\rangle=\sqrt{\hbar}|p=\hbar k\rangle$, while $dp=\hbar\,dk$. Therefore

$$\begin{aligned} |\Psi\rangle &=\int dk\, \hbar c_p(\hbar k)|p=\hbar k\rangle\\ &=\int dk\, \sqrt{\hbar}\,c_p(\hbar k)|k\rangle. \end{aligned}$$

Hence

$$c_k(k)=\sqrt{\hbar}\,c_p(\hbar k).$$

The norms agree because

$$\int dk\,\lvert c_k(k)\rvert^2 =\int dk\,\hbar\lvert c_p(\hbar k)\rvert^2 =\int dp\,\lvert c_p(p)\rvert^2.$$

3. A one-dimensional scattering state has unit incident amplitude, transmitted amplitude $t$, incident speed $v_i$, and transmitted speed $v_t$. Derive the transmission probability. How must $t$ be rescaled if both channel basis waves are normalized to unit flux?

Solution

A plane wave of amplitude $A$ carries current $j=v\lvert A\rvert^2$. Consequently,

$$T=\frac{j_t}{j_i} =\frac{v_t}{v_i}\lvert t\rvert^2.$$

A unit-flux incoming basis wave has spatial amplitude $1/\sqrt{v_i}$, and a unit-flux transmitted basis wave has amplitude $1/\sqrt{v_t}$. Writing the same solution in these bases gives the flux-normalized transmission amplitude

$$\widetilde t =\sqrt{\frac{v_t}{v_i}}\,t.$$

Therefore $T=\lvert\widetilde t\rvert^2$.

4. A nonuniform numerical grid uses positive quadrature weights $w_i$. Show how to convert sampled values $\psi_i$ satisfying $\sum_i w_i\lvert\psi_i\rvert^2=1$ into a vector $v$ with Euclidean norm one. Why does this conversion not by itself transform every Hamiltonian matrix into a Hermitian matrix?

Solution

Define

$$v_i=\sqrt{w_i}\,\psi_i.$$

Then

$$v^\dagger v =\sum_i\lvert v_i\rvert^2 =\sum_i w_i\lvert\psi_i\rvert^2 =1.$$

In matrix notation, $v=W^{1/2}\psi$ with $W=\operatorname{diag}(w_i)$. An operator represented by $H$ in the sampled-value coordinates transforms as

$$H_v=W^{1/2}HW^{-1/2}.$$

This matrix is Hermitian only if the original discretization is self-adjoint in the weighted inner product, equivalently $H^\dagger W=WH$. Rescaling coordinates cannot repair an inconsistent boundary stencil or a discretization that fails that condition.

5. If $u(r)=rR(r)$ and $\int_0^\infty \lvert u(r)\rvert^2\,dr=1$, show that the corresponding $R(r)$ is correctly radially normalized.

Solution

Since $u(r)=rR(r)$,

$$\lvert u(r)\rvert^2=\lvert R(r)\rvert^2r^2.$$

Therefore

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

References

• C. Cohen-Tannoudji, B. Diu, and F. Laloë, Quantum Mechanics, Wiley, 1977.
• D. J. Griffiths and D. F. Schroeter, Introduction to Quantum Mechanics, 3rd ed., Cambridge University Press, 2018.
• J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020.
• R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer, 1994.
• J. R. Taylor, Scattering Theory: The Quantum Theory of Nonrelativistic Collisions, Dover, 2006.