Weak Localization

Weak localization is the leading quantum-interference correction to diffusive conductivity when time-reversed scattering paths interfere constructively. It increases the probability of return, lowers the zero-field conductivity, and is suppressed by a magnetic field that breaks time-reversal coherence. The effect is “weak” because the correction is perturbative relative to the Drude conductivity, not because the phase coherence is poor.

Weak antilocalization is the symmetry-related sign reversal that commonly appears when spin–orbit coupling rotates spin along diffusive paths. Both phenomena occur on the metallic side,

$$k_{\mathrm F}\ell_{\mathrm{tr}}\gg1, \qquad g\gg1,$$

where $g$ is the appropriate dimensionless conductance. They are precursors to the scale dependence that becomes nonperturbative in Anderson Localization, but a low-field cusp is not itself proof of a strongly localized phase.

The calculation below follows the return probability through its momentum integral and magnetic-field dependence. We first use a single spin-degenerate band without spin–orbit or magnetic scattering, then explain how spin and other internal channels change the result.

Required background. Disorder in Quantum Matter supplies elastic scattering scales and the disorder ensemble. Anderson Localization supplies the distinction between coherent interference and classical diffusion. Quantum Coherence in Conductors supplies phase-memory lengths and the dephasing cutoffs used in the interference correction.

Coherent return paths and the diffusive regime

Take $e>0$ as the elementary charge and let the electron charge be $-e$. Define

$$\Delta\sigma(B) \equiv \sigma(B)-\sigma(0).$$

With this convention:

• ordinary weak localization gives $\Delta\sigma(B)>0$ at small nonzero perpendicular field;
• weak antilocalization gives $\Delta\sigma(B)<0$;
• for a weak Hall response and a small correction,

$$\Delta\rho_{xx}(B) \simeq -\rho_{xx}^2(0)\, \Delta\sigma_{xx}(B),$$

so the magnetoresistance sign is opposite to the magnetoconductance sign.

The diffusive calculation assumes a hierarchy such as

$$k_{\mathrm F}^{-1} \ll \ell_{\mathrm{tr}} \ll L_\phi,$$

and fields low enough that the elastic-scale orbital dynamics, Landau quantization of the electronic spectrum, and classical multiband magnetoresistance do not replace the diffusive interference problem. A useful transport field is

$$B_{\mathrm{tr}} \equiv \frac{\hbar}{4eD\tau_{\mathrm{tr}}},$$

up to convention-dependent factors connecting $D\tau_{\mathrm{tr}}$ to $\ell_{\mathrm{tr}}^2$. A simple Hikami–Larkin–Nagaoka fit should remain well below this scale.

Time-Reversed Paths

Consider a closed diffusive path $\mathcal P$ that begins and ends near the same point. In a time-reversal-invariant spinless problem, its amplitude and that of the reversed path $\bar{\mathcal P}$ are

$$\mathcal A_{\mathcal P} = \lvert A_{\mathcal P}\rvert e^{i\phi_{\mathcal P}}.$$ $$\mathcal A_{\bar{\mathcal P}} = \lvert A_{\mathcal P}\rvert e^{i\phi_{\bar{\mathcal P}}},$$

with

$$\phi_{\mathcal P} = \phi_{\bar{\mathcal P}} \pmod{2\pi}$$

at zero magnetic field. Their return probability is therefore

$$\begin{aligned} \left| \mathcal A_{\mathcal P} + \mathcal A_{\bar{\mathcal P}} \right|^2 &= \lvert\mathcal A_{\mathcal P}\rvert^2 + \lvert\mathcal A_{\bar{\mathcal P}}\rvert^2 + 2\operatorname{Re} \left( \mathcal A_{\mathcal P} \mathcal A_{\bar{\mathcal P}}^\ast \right) \\ &= 4\lvert A_{\mathcal P}\rvert^2. \end{aligned}$$

An incoherent sum would give only $2\lvert A_{\mathcal P}\rvert^2$. The enhanced return probability removes weight from long-range diffusion and lowers conductivity.

Most unrelated path pairs acquire random relative phases and disappear under disorder averaging. The path and its exact reverse survive because each visits the same impurities in the opposite order. This is why weak localization is an ensemble-averaged correction even though Universal Conductance Fluctuations remain sample-specific.

Two consequences of time reversal. The upper panel shows a closed loop traversed in opposite directions; the arrows label the two traversals of the same loop. At zero field their phases agree and enhance return. Perpendicular flux creates a relative phase $4\pi\Phi/\Phi_0$ and suppresses the Cooperon. In the stated convention $\Delta\sigma(B)=\sigma(B)-\sigma(0)$, weak localization bends upward and weak antilocalization bends downward. The lower curves illustrate these signs, not a fit or a quantitatively specified spin-channel model.

Quantum Correction to Conductivity

Return probability

The diffusion kernel in $d$ dimensions is

$$P(\mathbf r,t) = \frac{1}{(4\pi Dt)^{d/2}} \exp\!\left( -\frac{r^2}{4Dt} \right).$$

The return probability density is

$$P(\mathbf 0,t) = \frac{1}{(4\pi Dt)^{d/2}}.$$

Interference accumulates after motion becomes diffusive, $t\gtrsim\tau_{\mathrm{tr}}$, and is cut off after phase memory is lost, $t\gtrsim\tau_\phi$. The correction therefore has the structural form

$$\delta\sigma \propto -\frac{e^2D}{\hbar} \int_{\tau_{\mathrm{tr}}}^{\tau_\phi} dt\, P(\mathbf0,t).$$

In two dimensions, $P(\mathbf0,t)\propto1/t$, so every logarithmic interval of return time contributes comparably:

$$\delta\sigma_{2\mathrm D} \propto -\ln \left( \frac{\tau_\phi}{\tau_{\mathrm{tr}}} \right).$$

Long return paths make the correction sensitive to phase memory over many elastic collisions.

Cooperon mode integral

After disorder averaging, the path–reverse-path interference is represented by the Cooperon. In the simplest scalar channel,

$$C(\mathbf q,\omega) = \frac{1}{ Dq^2-i\omega+\tau_\phi^{-1} }.$$

For one spin-degenerate band with no valley degeneracy, spin–orbit coupling or magnetic scattering, the zero-frequency correction is

$$\delta\sigma = - \frac{2e^2D}{\pi\hbar} \int \frac{d^dq}{(2\pi)^d} \frac{1}{ Dq^2+\tau_\phi^{-1} },$$

The prefactor includes two equivalent spin copies. A strictly spinless orthogonal model gives half this correction. Spin degeneracy is a physical factor, not a cutoff convention; the singlet/triplet description below explains how the prefactor changes with spin relaxation.

The ultraviolet cutoff is $q_{\max}\sim\ell_{\mathrm{tr}}^{-1}$. Dephasing already regularizes the infrared through $\tau_\phi^{-1}$ in the denominator. A finite sample replaces the integral by its diffusion modes, including escape to leads. Magnetic field and spin relaxation alter those modes as described below. A hard lower cutoff $q_{\min}\sim1/L_\phi$ is an alternative logarithmic approximation, not an additional independent dephasing mechanism.

For a large two-dimensional sample with the stated two-spin convention,

$$\delta\sigma_{2\mathrm D}(0) \simeq - \frac{e^2}{2\pi^2\hbar} \ln \left( \frac{\tau_\phi}{\tau_{\mathrm{tr}}} \right)$$

up to finite constants absorbed into the short-scale conductivity. Equivalently,

$$\delta\sigma_{2\mathrm D}(0) \simeq - \frac{e^2}{\pi^2\hbar} \ln \left( \frac{L_\phi}{\ell_{\mathrm{tr}}} \right)$$

when the length and time cutoffs are matched at logarithmic accuracy.

The correction must remain perturbative:

$$\left| \delta\sigma \right| \ll \sigma_{\mathrm D}.$$

Once it becomes comparable with the Drude value, a first-order weak-localization formula is not self-consistent.

Dimensional dependence

The same return integral behaves differently with dimension:

Effective dimension	Dependence on coherence length
Quasi-1D long wire	$\delta G\propto-L_\phi/L$ for $L\gg L_\phi$
2D film	$\delta\sigma\propto-\ln(L_\phi/\ell)$
3D bulk	Short-scale constant plus a term proportional to $+1/L_\phi$

The effective dimension is set by the diffusion length over the relevant time. A film is two-dimensional for the interference correction when

$$t_{\mathrm{film}} \ll L_\phi$$

and the magnetic length over the fitted field range also exceeds its thickness. A narrow wire becomes quasi-one-dimensional when both transverse dimensions are smaller than the active coherence scale. The thermal averaging length $L_T=\sqrt{\hbar D/(k_{\mathrm B}T)}$ is not an additional universal cutoff for this ensemble-averaged WL correction. It matters separately for conductance fluctuations and interaction corrections. Temperature still affects WL through dephasing and any energy dependence of the transport parameters; see Beenakker and van Houten (1991).

Magnetic-Field Suppression

The Cooperon sees charge twice

For a closed path enclosing flux

$$\Phi = \oint_{\mathcal P} \mathbf A\cdot d\mathbf l,$$

the orbital phase changes sign when the path is reversed. Choosing the reversed path minus the forward path, the relative phase is

$$\Delta\phi_B = \frac{2e}{\hbar}\Phi = 4\pi \frac{\Phi}{\Phi_0}.$$

Here $\Phi_0=h/e$. The factor $2e$ is not a Cooper-pair charge. It arises because the interference compares one electron path with its reverse. In the Cooperon equation, the gradient is correspondingly replaced by

$$-i\nabla \longrightarrow -i\nabla - \frac{2e}{\hbar}\mathbf A.$$

A perpendicular field quantizes the diffusive transverse modes:

$$Dq^2 \longrightarrow \frac{4eD\lvert B\rvert}{\hbar} \left( n+\frac12 \right).$$

The field scale associated with a decay time $\tau_i$ is

$$B_i = \frac{\hbar}{4eD\tau_i} = \frac{\hbar}{4eL_i^2}.$$

Here $L_i=\sqrt{D\tau_i}$. In particular,

$$B_\phi = \frac{\hbar}{4eD\tau_\phi} = \frac{\hbar}{4eL_\phi^2}.$$

Long coherent loops enclose larger areas and are destroyed by smaller fields.

Two-dimensional line shape

Define

$$\mathcal F(z) = \psi \left( \frac12+\frac{1}{z} \right) + \ln z, \qquad z>0,$$

where $\psi$ is the digamma function and $\mathcal F(0)=0$ is defined by continuity. Summing the field-quantized diffusion modes gives the digamma function. The low-field single-scale limit of Hikami, Larkin and Nagaoka (1980) can be written

$$\Delta\sigma(B) = \alpha \frac{e^2}{2\pi^2\hbar} \mathcal F \left( \frac{\lvert B\rvert}{B_\phi} \right).$$

For the two-spin orthogonal example above, $\alpha=1$; for its spinless version, $\alpha=1/2$. A different internal-channel problem needs its own prefactor and relaxation scales.

At small field,

$$\mathcal F(z) = \frac{z^2}{24} - \frac{7z^4}{960} + O(z^6),$$

so the ideal orbital curve is even and quadratic at the origin. In the present sign convention:

$$\begin{array}{lll} \alpha>0 &\Rightarrow& \text{weak localization},\\[3pt] \alpha<0 &\Rightarrow& \text{weak antilocalization}. \end{array}$$

Other papers define $\Delta\sigma$ with the opposite order or place a minus sign in front of $\alpha$. Always read the equation before interpreting the fitted sign.

Worked field scale

Let

$$D = 0.020\ \mathrm{m^2\,s^{-1}}.$$ $$\tau_\phi = 50\ \mathrm{ps}.$$

Then

$$L_\phi = \sqrt{D\tau_\phi} = 1.00\ \mu\mathrm m$$

and

$$B_\phi = \frac{\hbar}{4eL_\phi^2} \simeq 0.165\ \mathrm{mT}.$$

This scale sets the crossover near the origin; the digamma curve has no sharp outer width. A different measured crossover can indicate a different coherent length or additional relaxation channels.

Weak Antilocalization

For spin-$1/2$ particles, time reversal satisfies

$$\mathcal T^2=-1.$$

Spin–orbit coupling rotates spin as momentum changes. The reversed trajectory retraces the orbital path but also carries a constrained spin rotation. In the symplectic regime, the net path–reverse-path interference can reduce return probability rather than enhance it. The conductivity correction is then positive at zero field, and suppressing it with field gives negative $\Delta\sigma(B)$ in the convention used here.

This is weak antilocalization. It remains a small interference correction:

$$\left| \delta\sigma_{\mathrm{WAL}} \right| \ll \sigma_{\mathrm D}.$$

The name does not mean that all states are extended, nor that strong disorder cannot eventually localize the system. It identifies the sign of the perturbative quantum correction in a symmetry regime.

A $\pi$ Berry phase can provide an equivalent language in some Dirac systems, but a WAL cusp is not unique evidence for a topological surface state. Rashba-split semiconductor bands, ordinary heavy-element films, coupled quantum wells, graphene valley structure, and several bulk channels can all produce WAL-like magnetoconductance.

Spin–Orbit and Internal Channels

Singlet and triplet Cooperons

The product of two spin-$1/2$ amplitudes decomposes into one singlet and three triplet Cooperon modes:

$$\frac12\otimes\frac12 = 0\oplus1.$$

Without spin relaxation, the channel sum produces the ordinary orthogonal correction. Spin–orbit coupling relaxes or mixes the triplet modes while the singlet has the opposite interference sign. In the strong spin–orbit limit, the surviving low-energy balance is antilocalizing.

A useful schematic channel formula is

$$\Delta\sigma(B) = \frac{e^2}{2\pi^2\hbar} \sum_a \alpha_a \mathcal F \left[ \frac{\lvert B\rvert}{ B_\phi+B_a } \right],$$

where

$$B_a = \frac{\hbar\Gamma_a}{4eD}$$

encodes a channel relaxation rate $\Gamma_a$. The coefficients $\alpha_a$ and the combinations entering $B_a$ must come from the microscopic symmetry problem. Replacing them by several unconstrained copies of one curve can fit noise while obscuring the physics.

In the simplified strong-spin–orbit two-dimensional convention, one independent coherent symplectic channel often contributes approximately

$$\alpha \simeq -\frac12.$$

This is a benchmark, not a universal channel counter. Interchannel scattering, unequal diffusion constants, magnetic scattering, finite thickness, Zeeman mixing, and partial coherence shift the fitted value.

Spin-relaxation mechanisms

For Elliott–Yafet relaxation, scattering with spin admixture produces spin flips. If $b^2$ is a representative spin-mixing probability and $\tau_c$ the relevant collision time, the schematic estimate is

$$\frac{1}{\tau_{\mathrm{so}}} \sim \frac{b^2}{\tau_c}.$$

More frequent momentum scattering then causes faster spin relaxation.

For Dyakonov–Perel relaxation, carriers precess in a momentum-dependent spin–orbit field $\boldsymbol\Omega(\mathbf k)$ between collisions. In the motional-narrowing regime, $\sqrt{\langle\Omega^2\rangle}\tau_c\ll1$,

$$\frac{1}{\tau_{\mathrm{so}}} \sim \left\langle \Omega^2 \right\rangle \tau_c.$$

Here $\tau_c$ is the correlation time of the spin–orbit field. More frequent momentum scattering interrupts precession and can lengthen the spin lifetime. For isotropic simple models $\tau_c$ can be comparable to $\tau_{\mathrm{tr}}$, but anisotropic scattering and different angular harmonics need distinct times. The Cooperon spin-relaxation rates also need not equal a measured qubit $T_2^{-1}$. Gate-dependent Rashba coupling, Dresselhaus anisotropy, cubic terms, and several occupied subbands can require an Iordanskii–Lyanda-Geller–Pikus or related model rather than a scalar Hikami–Larkin–Nagaoka fit.

Valley, surface, and band structure

Internal degrees of freedom create additional interference channels:

• intervalley scattering in graphene can restore ordinary weak localization by coupling valleys;
• intravalley chirality and trigonal warping gap different graphene Cooperons;
• top and bottom surfaces of a thin topological insulator can behave as two channels only if their coherent coupling is weak;
• bulk and surface carriers with different $D$ and $\tau_\phi$ cannot generally be represented by one common $B_\phi$;
• magnetic order or magnetic impurities break time reversal and can suppress both WL and WAL.

The fitted $\alpha$ is therefore a compressed summary of a model, not a direct measurement of “the number of topological channels.”

From Measured Resistance to Conductivity

Convert measured resistances to sheet resistivities using the device geometry. The resulting tensor must be inverted before fitting a conductivity formula:

$$\boldsymbol\sigma = \boldsymbol\rho^{-1}.$$

For an isotropic two-dimensional Hall bar,

$$\sigma_{xx} = \frac{\rho_{xx}}{ \rho_{xx}^2+\rho_{xy}^2 }.$$ $$\sigma_{xy} = -\frac{\rho_{xy}}{ \rho_{xx}^2+\rho_{xy}^2 }.$$

If Hall mixing and relative corrections are negligible, the first-order conversion $\Delta\sigma_{xx}\simeq-\Delta\rho_{xx}/\rho_{xx}^2(0)$ can replace the full inverse. The raw resistance change is still not a conductivity change.

For a nonmagnetic sample with reversible field sweeps, longitudinal symmetrization reduces odd Hall admixture from misaligned contacts:

$$\rho_{xx}^{\mathrm{sym}}(B) = \frac{ \rho_{xx}^{\mathrm{raw}}(B) + \rho_{xx}^{\mathrm{raw}}(-B) }{2},$$

The corresponding Hall antisymmetrization is:

$$\rho_{xy}^{\mathrm{asym}}(B) = \frac{ \rho_{xy}^{\mathrm{raw}}(B) - \rho_{xy}^{\mathrm{raw}}(-B) }{2}.$$

Check reciprocity and field history before using these parity projections: hysteresis or magnetic order can invalidate them. A field offset can also reflect trapped flux rather than an arbitrary shift of the curve.

What a magnetoconductance fit can determine

A useful fit starts with a diffusive field range, the correct sample dimension and a conductivity tensor. Vary the maximum fitted field to see whether $B_\phi$ and the prefactor remain stable. Background curvature, channel prefactors and relaxation fields can compensate one another; inspect their joint uncertainty rather than trusting each fitted number separately.

Temperature, gate voltage and field orientation provide independent comparisons. They can change density, occupied subbands and diffusion constants as well as dephasing. In a thin film, the orbital correction primarily follows the perpendicular field; a parallel-field response may instead reveal finite thickness, Zeeman coupling or magnetic scattering. Interaction corrections and superconducting fluctuations have their own dependencies and should not automatically be absorbed into an HLN prefactor.

Finally, verify the electron temperature and current dependence before interpreting low-temperature saturation of $L_\phi$. A comparison with conductance fluctuations or ring oscillations can test the inferred coherence scale, provided their different thermal averaging and geometric sensitivities are included.

Exercises

1. Two-dimensional logarithm

Evaluate

$$I_{2\mathrm D} = \int_{1/L_\phi}^{1/\ell} \frac{d^2q}{(2\pi)^2} \frac{1}{Dq^2}.$$

Use it to recover the logarithmic weak-localization correction.

Solution

In polar coordinates,

$$\begin{aligned} I_{2\mathrm D} &= \frac{1}{(2\pi)^2} \int_0^{2\pi}d\theta \int_{1/L_\phi}^{1/\ell} \frac{q\,dq}{Dq^2}\\ &= \frac{1}{2\pi D} \ln \left( \frac{L_\phi}{\ell} \right). \end{aligned}$$

Substituting into the two-spin orthogonal correction gives

$$\begin{aligned} \delta\sigma &=-\frac{2e^2D}{\pi\hbar}I_{2\mathrm D}\\ &=-\frac{e^2}{\pi^2\hbar} \ln \left( \frac{L_\phi}{\ell} \right). \end{aligned}$$

Since $L_\phi^2=D\tau_\phi$ and $\ell^2$ is proportional to $D\tau_{\mathrm{tr}}$ up to a dimensional factor,

$$\delta\sigma \simeq - \frac{e^2}{2\pi^2\hbar} \ln \left( \frac{\tau_\phi}{\tau_{\mathrm{tr}}} \right)$$

at logarithmic accuracy.

2. Magnetic phase of a reversed loop

Show that the orbital phase difference between a closed electron path and its reverse is $2e\Phi/\hbar$. What flux gives a phase difference of order one?

Solution

For electron charge $-e$, the vector-potential phase is

$$\phi_{\mathcal P}^{(B)} = -\frac{e}{\hbar} \oint_{\mathcal P} \mathbf A\cdot d\mathbf l = -\frac{e\Phi}{\hbar}.$$

Reversing the path reverses the line integral:

$$\phi_{\bar{\mathcal P}}^{(B)} = +\frac{e\Phi}{\hbar}.$$

Their difference has magnitude

$$\lvert\Delta\phi_B\rvert = \frac{2e\lvert\Phi\rvert}{\hbar} = 4\pi \frac{\lvert\Phi\rvert}{\Phi_0}.$$

Suppression begins when this is of order one, so the relevant flux is of order

$$\lvert\Phi\rvert \sim \frac{\hbar}{2e} = \frac{\Phi_0}{4\pi}.$$

Exact line-shape factors come from averaging the distribution of loop areas, not from setting one sharp flux threshold.

3. Small-field curvature

Use the large-argument expansion of the digamma function to show

$$\mathcal F(z) = \frac{z^2}{24} + O(z^4).$$

What is $\Delta\sigma/(e^2/h)$ at $\lvert B\rvert=0.2B_\phi$ for $\alpha=1$ at leading order?

Solution

For $x\gg1$,

$$\psi \left( x+\frac12 \right) = \ln x + \frac{1}{24x^2} + O(x^{-4}).$$

Set $x=1/z$. Then

$$\mathcal F(z)=\psi\left(\frac12+\frac1z\right)+\ln z =\frac{z^2}{24}+O(z^4).$$

At $z=0.2$,

$$\mathcal F(0.2) \simeq \frac{0.2^2}{24} = 1.667\times10^{-3}.$$

Because

$$\frac{ e^2/(2\pi^2\hbar) }{ e^2/h } = \frac1\pi,$$

the leading correction is

$$\frac{\Delta\sigma}{e^2/h} \simeq \frac{\mathcal F(0.2)}{\pi} \simeq 5.31\times10^{-4}.$$

4. Dimensional crossover

A film is $30\ \mathrm{nm}$ thick. Its fitted $L_\phi$ decreases from $600\ \mathrm{nm}$ at $1\ \mathrm K$ to $20\ \mathrm{nm}$ at $40\ \mathrm K$. Can one use one two-dimensional formula throughout?

Solution

At $1\ \mathrm K$,

$$t_{\mathrm{film}} \ll L_\phi,$$

so transverse diffusion samples the full thickness before phase memory is lost; only the lowest transverse diffusion mode remains long lived. A two-dimensional orbital treatment can be appropriate.

At $40\ \mathrm K$,

$$L_\phi < t_{\mathrm{film}},$$

so a coherent path need not reach the thickness boundaries. It samples locally three-dimensional diffusion, and a two-dimensional digamma fit is no longer justified without a finite-thickness calculation. The crossover may occur before $L_\phi=t_{\mathrm{film}}$ if the magnetic length becomes the shorter resolving scale. The separate thermal averaging length is not a universal WL thickness cutoff.

5. Coupled WAL channels

Two equivalent, independent symplectic surfaces each have $\alpha=-1/2$. What prefactor is expected if they are independent? What can happen when coherent intersurface scattering is much faster than dephasing?

Solution

Independent corrections add:

$$\alpha_{\mathrm{tot}} = -\frac12-\frac12 = -1.$$

If coherent intersurface scattering is fast, the two surface labels are not separately conserved over $\tau_\phi$. One symmetric long-lived Cooperon combination can remain while the orthogonal combination is gapped. The low-field response can then approach

$$\alpha_{\mathrm{tot}} \simeq -\frac12.$$

Intermediate coupling, unequal $D$, bulk mediation, or magnetic scattering gives nonuniversal values. Thus observing $\alpha=-1$ or $-1/2$ does not by itself prove two or one topological surfaces.

Related calculations

• Scaling Theory of Localization turns the perturbative logarithm into a length-dependent conductance flow.
• Quantum Coherence in Conductors explains microscopic dephasing and the distinction from thermal averaging.
• Graphene and Dirac Materials introduces the valley and chirality structure behind graphene interference.
• Kubo Formula develops the linear-response framework underlying the conductivity correction.

References

• C. W. J. Beenakker and H. van Houten, “Quantum Transport in Semiconductor Nanostructures,” Solid State Physics 44, 1–228 (1991), doi:10.1016/S0081-1947(08)60091-0. Provides mesoscopic transport, coherence, and geometry context.
• G. Bergmann, “Weak Localization in Thin Films: A Time-of-Flight Experiment with Conduction Electrons,” Physics Reports 107, 1–58 (1984), doi:10.1016/0370-1573(84)90103-0. Classic physical and experimental review.
• S. Hikami, A. I. Larkin, and Y. Nagaoka, “Spin–Orbit Interaction and Magnetoresistance in the Two Dimensional Random System,” Progress of Theoretical Physics 63, 707–710 (1980), doi:10.1143/PTP.63.707. Derives the canonical two-dimensional field and spin–orbit structure.
• S. V. Iordanskii, Yu. B. Lyanda-Geller, and G. E. Pikus, “Weak Localization in Quantum Wells with Spin–Orbit Interaction,” JETP Letters 60, 206–210 (1994). Develops spin-precession channel structure beyond a scalar spin-flip time.
• W. Knap, C. Skierbiszewski, A. Zduniak, et al., “Weak Antilocalization and Spin Precession in Quantum Wells,” Physical Review B 53, 3912–3924 (1996), doi:10.1103/PhysRevB.53.3912. Tests spin–orbit interference in semiconductor wells.
• P. A. Lee and T. V. Ramakrishnan, “Disordered Electronic Systems,” Reviews of Modern Physics 57, 287–337 (1985), doi:10.1103/RevModPhys.57.287. Reviews localization, interactions, and transport corrections.
• E. McCann, K. Kechedzhi, V. I. Fal’ko, H. Suzuura, T. Ando, and B. L. Altshuler, “Weak-Localization Magnetoresistance and Valley Symmetry in Graphene,” Physical Review Letters 97, 146805 (2006), doi:10.1103/PhysRevLett.97.146805. Resolves graphene valley and chirality channels.