Tilt-Corrected Differential Phase Contrast (tcDPC)

• tcDPC is a 4D-STEM technique that corrects real-space shifts to isolate antisymmetric phase-gradient information.
• It subtracts tilt-corrected conjugate detector pairs under the weak phase object approximation to recover the imaginary phase component.
• The method is non-iterative and dose-efficient, offering rapid imaging though it omits symmetric low-frequency contrast.

Tilt-corrected differential phase contrast (tcDPC) is a 4D-STEM imaging mode in which the antisymmetric, odd-in-detector-coordinate component of the bright-field (BF) disk is combined after correcting the real-space shifts induced by defocus and other probe-forming aberrations. In the weak phase object approximation (WPOA), tcDPC isolates the imaginary, antisymmetric part of the phase-contrast transfer and yields a real-space differential phase image. It belongs to a broader class of tilt-corrected direct ptychographic methods that also includes tilt-corrected bright-field (tcBF) and aberration-corrected bright-field (acBF), the latter combining the symmetric and antisymmetric channels to obtain continuously nonzero transfer up to the BF information limit of $2\alpha$ (Ma et al., 1 Oct 2025, Ma et al., 28 Jul 2025).

1. Definition and physical basis

In 4D-STEM, a convergent-beam electron diffraction pattern is recorded at each probe position, so the dataset may be regarded as a collection of virtual BF STEM images, one for each detector pixel inside the BF disk. Because each detector pixel subtends a very small angle for a convergent probe, each virtual BF image is highly coherent and encodes phase-contrast information efficiently. Defocus and other aberrations introduce detector-dependent real-space “parallax” shifts, so a direct sum over detector pixels blurs the coherent information unless those shifts are removed (Ma et al., 1 Oct 2025).

tcDPC is formed by subtracting tilt-corrected conjugate, or Friedel, pairs of virtual BF images at $\pm\boldsymbol{\Theta}$ over the BF disk. Under WPOA and defocus-only conditions, this subtraction isolates the imaginary, antisymmetric part of the complex phase-contrast transfer, producing a real-space gradient image. The complementary tcBF operation instead sums the same shift-corrected images and recovers the real, symmetric part of the transfer. In the generalized contrast description revisiting Rose’s formalism, the antisymmetric DPC-like phase contrast is associated with the double-overlap (DO) region of the BF disk, whereas the symmetric phase contrast is associated with the triple-overlap (TO) region (Ma et al., 28 Jul 2025).

The physical motivation for tilt correction is simple in the defocus-only case. The virtual image associated with detector coordinate $\boldsymbol{\Theta}$ is shifted by

$\Delta\rho=\Delta f\,\boldsymbol{\Theta},$

equivalently by the Fourier phase factor $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$, where $k_0=2\pi/\lambda$. More generally, for arbitrary aberration phase $\chi$, the shift becomes

$$\Delta\rho(\boldsymbol{\Theta})=\nabla\chi(\boldsymbol{\Theta}).$$

Tilt correction removes these shifts and thereby restores coherence across the set of virtual BF images before symmetric or antisymmetric combination (Ma et al., 1 Oct 2025).

2. Transfer theory under WPOA

For a single detector pixel at position $\boldsymbol{\Theta}$, the complex phase-contrast transfer function (PCTF) under WPOA is written as

$$\operatorname{PCTF}(\boldsymbol{\omega}, \boldsymbol{\Theta})=\frac{i}{2 \Omega_0} A(\boldsymbol{\Theta})\left\{A(\boldsymbol{\omega}-\boldsymbol{\Theta}) e^{-i[\chi(\boldsymbol{\omega}-\boldsymbol{\Theta})-\chi(\boldsymbol{\Theta})]}-A(\boldsymbol{\omega}+\boldsymbol{\Theta}) e^{+i[\chi(\boldsymbol{\omega}+\boldsymbol{\Theta})-\chi(\boldsymbol{\Theta})]}\right\},$$

where $\pm\boldsymbol{\Theta}$0 is spatial frequency, $\pm\boldsymbol{\Theta}$1 is the illumination solid angle, and $\pm\boldsymbol{\Theta}$2 is the aperture function inside the BF disk $\pm\boldsymbol{\Theta}$3 (Ma et al., 1 Oct 2025).

For defocus only,

$\pm\boldsymbol{\Theta}$4

and the detector-pair symmetry

$\pm\boldsymbol{\Theta}$5

implies that summation over $\pm\boldsymbol{\Theta}$6 cancels the imaginary part, whereas subtraction cancels the real part. Denoting by $\pm\boldsymbol{\Theta}$7 the normalized overlap area of two BF disks separated by $\pm\boldsymbol{\Theta}$8, the recovered tcDPC transfer is

$\pm\boldsymbol{\Theta}$9

The corresponding tcBF transfer is

$\boldsymbol{\Theta}$0

Thus tcDPC is purely imaginary and antisymmetric, whereas tcBF is real-valued and symmetric (Ma et al., 1 Oct 2025).

In focus, $\boldsymbol{\Theta}$1, tcDPC reaches its maximal transfer and matches the information transfer of in-focus single-sideband ptychography in the noise-free limit. Because tcDPC is a differential, or gradient, contrast, an integrated form can be constructed in Fourier space by dividing by $\boldsymbol{\Theta}$2; the papers denote this as i-tcDPC and describe it as analogous to iDPC processing, including the need for regularization at low spatial frequency (Ma et al., 28 Jul 2025).

3. Reconstruction workflow and implementation

The practical tcDPC pipeline starts with a 4D-STEM acquisition using a pixelated direct electron detector that records the BF disk at each scan position. The BF disk radius is set by the probe semi-convergence angle $\boldsymbol{\Theta}$3, and the detector sampling must be sufficiently fine; one formulation recommends detector pixel angular size $\boldsymbol{\Theta}$4 to $\boldsymbol{\Theta}$5 to maintain coherence of the per-pixel virtual images. The scan step may be coarser than the Nyquist sampling for $\boldsymbol{\Theta}$6, with reconstruction upsampling such as $\boldsymbol{\Theta}$7 to $\boldsymbol{\Theta}$8 used to achieve finer spatial sampling (Ma et al., 28 Jul 2025, Ma et al., 1 Oct 2025).

Before forming tcDPC, the BF disk is recentered to remove residual probe tilt. A robust estimate is the scan-averaged BF center of mass, yielding a disk shift $\boldsymbol{\Theta}$9; the diffraction patterns are then recentered by defining $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$0. Within the BF mask, one may explicitly form the symmetric and antisymmetric diffraction components

$\Delta\rho=\Delta f\,\boldsymbol{\Theta},$1

with tcDPC arising from the antisymmetric component after parallax correction (Ma et al., 28 Jul 2025).

Shift estimation is then performed across detector coordinates. For defocus-only data, the shift law is $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$2. For general aberrations, the shift is estimated as $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$3, typically by cross-correlation of nearby detector-pixel images followed by fitting a smooth aberration surface. Each virtual BF image is corrected either by real-space registration or by multiplication with the Fourier phase ramp $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$4 (Ma et al., 1 Oct 2025).

tcDPC is then constructed by subtracting conjugate pairs of shift-corrected images at $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$5 and $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$6. Equivalent practical forms include an opposing-pixel subtraction and an odd-in-$\Delta\rho=\Delta f\,\boldsymbol{\Theta},$7 first moment over the tilt-corrected BF disk. Optional normalization by the total BF current removes dose variations, and optional i-tcDPC divides by $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$8 in Fourier space to estimate a phase image. These operations are non-iterative and computationally light compared with iterative ptychography (Ma et al., 28 Jul 2025, Ma et al., 1 Oct 2025).

4. Aberrations, higher-order distortions, and Scherzer defocus

The central theoretical complication beyond defocus is that higher-order aberrations break the simple conjugate symmetry that underlies the clean tcBF/tcDPC separation. Although the detector-dependent image shift still follows $\Delta\rho=\Delta f\,\boldsymbol{\Theta},$9, additional cross-terms in $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$0 generate frequency-dependent phase modulations, described in the paper as dispersive shifts. As a result, simple tilt correction by real-space translation is no longer sufficient; per-image frequency-dependent phase correction is required to preserve coherence (Ma et al., 1 Oct 2025).

For spherical aberration $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$1, the single-pixel PCTF acquires a distortion term

$$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$2

which is independent of defocus and grows toward the aperture edge. In the TO region, the effective aberration inside the sine term includes a detector-dependent increase in defocus by $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$3, an induced first-order astigmatism term $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$4, and the original spherical term. These detector-dependent effective aberrations cause strong damping if the virtual BF images are summed without proper per-image correction (Ma et al., 1 Oct 2025).

For two-fold astigmatism $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$5, the effective defocus becomes direction-dependent: along one axis, defocus and $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$6 add; along the orthogonal axis, they subtract. The paper states that tcBF shows no contrast along the zero-defocus axis, and acBF recovers only the antisymmetric scattering there, implying irrecoverable loss of symmetric information in that direction. A plausible implication is that tcDPC is comparatively robust precisely because it targets the antisymmetric component, but it does not restore the missing symmetric information (Ma et al., 1 Oct 2025).

A special case arises at Scherzer defocus in a spherically-aberration-limited system,

$$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$7

Under these conditions, the probe-forming lens phase approximates a phase plate, bringing scattered beams closer to a Zernike-like $$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$8 phase condition over a wide angular range. The papers argue that, after per-image correction, this smooths oscillations in the acBF transfer and dose efficiency. For tcDPC specifically, the relevant point is that tilt-corrected antisymmetric transfer remains available, but the full benefit of the phase-plate-like behavior appears only when tcDPC is combined with tcBF in acBF (Ma et al., 1 Oct 2025).

5. Relation to tcBF, acBF, conventional DPC, and ptychography

tcDPC is one member of a symmetric/antisymmetric decomposition of BF information. tcBF is obtained by summing tilt-corrected virtual BF images over the BF disk and recovers the real, symmetric component of the transfer. tcDPC is obtained by subtracting conjugate detector pairs and recovers the imaginary, antisymmetric component. The combination of the two, after per-image aberration correction, yields acBF, whose defocus-only analytic transfer is

$$e^{-i\,k_0(\Delta f\,\boldsymbol{\Theta})\cdot\boldsymbol{\omega}}$$9

Because tcBF has a $k_0=2\pi/\lambda$0 modulation and tcDPC has a $k_0=2\pi/\lambda$1 modulation with complementary envelopes, the two channels do not share simultaneous zeros. The stated consequence is continuously nonzero transfer to $k_0=2\pi/\lambda$2 after proper phase correction (Ma et al., 1 Oct 2025).

Relative to conventional segmented-detector DPC, tcDPC is a fully pixelated 4D-STEM generalization that uses the full BF disk and explicitly corrects defocus-induced parallax. Relative to in-focus iDPC and in-focus single-sideband ptychography, tcDPC shares the emphasis on antisymmetric scattering and, at focus, the same information transfer as in-focus SSB in the noise-free limit. What it does not provide is symmetric low-frequency transfer; that is the role of defocus-enabled tcBF, and acBF combines both regimes into a directly interpretable phase-contrast image (Ma et al., 28 Jul 2025, Ma et al., 1 Oct 2025).

The papers therefore place tcDPC within direct, non-iterative ptychography rather than iterative phase retrieval. Tilt-corrected methods “re-map” detector pixels to image pixels using known shifts and phase corrections, whereas iterative multislice or full-field ptychography is computationally much more intensive but can exceed the $k_0=2\pi/\lambda$3 diffraction limit and handle multiple scattering. This suggests that tcDPC occupies a specific methodological niche: a fast, linear, and dose-efficient route to the antisymmetric BF information, especially valuable when iterative reconstruction is unnecessary or impractical (Ma et al., 28 Jul 2025).

6. Dose efficiency, empirical behavior, and limitations

For weakly scattering samples and a detector with $k_0=2\pi/\lambda$4 and approximately flat noise power spectrum,

$k_0=2\pi/\lambda$5

Within this framework, tcBF suffers from oscillatory transfer and from strong damping under higher-order aberrations, whereas tcDPC is maximized in focus and is dose-efficient for antisymmetric scattering. The principal limitation of tcDPC is that it lacks symmetric low-frequency transfer. The acBF combination is therefore presented as having smoother, continuously nonzero DQE to $k_0=2\pi/\lambda$6 and improved overall dose efficiency relative to tcBF or tcDPC alone (Ma et al., 1 Oct 2025).

The comparative demonstrations reported in the papers are consistent with that division of labor. In a simulated single atom at 300 kV with 30 mrad convergence, $k_0=2\pi/\lambda$7m, and Scherzer defocus, tcBF showed poorer resolution and long negative tails due to summation of images with different effective defocus and astigmatism, whereas acBF gave a sharp point response without delocalizing tails. In a simulated MOF NU-1000 dataset at 300 kV, 10 mrad, $k_0=2\pi/\lambda$8 mm, Scherzer defocus, $k_0=2\pi/\lambda$9, 1 Å scan, and upsampling $\chi$0, tcBF improved large-feature contrast but had poorer fine resolution than axial BF, while acBF resolved heavy Zr atoms in linkers. In experimental WSe$\chi$1 bilayer data at 80 kV and 25 mrad in an aberration-corrected system, tcBF remained oscillatory and showed strong contrast reversals, whereas acBF yielded interpretable phase contrast with correct atomic positions and an information limit approaching $\chi$2 (Ma et al., 1 Oct 2025).

These results do not imply that tcDPC is superseded. Rather, they delimit its role. tcDPC is robust in focus, non-iterative, and dose-efficient for the antisymmetric channel; in-focus tcDPC also matches in-focus SSB transfer in the noise-free limit. Its limitations are structural rather than algorithmic: it is a differential phase signal and therefore does not carry the symmetric low-frequency transfer that tcBF and acBF recover with defocus (Ma et al., 28 Jul 2025).

The formal derivations also impose clear assumptions. The linear transfer interpretation relies on WPOA, thin weak-phase samples, and small amplitude modulation. Finite detector pixel size reduces coherence if it is not small compared with $\chi$3. Residual tilt and uncorrected higher-order aberrations degrade the reconstruction, and multiple or inelastic scattering introduces coherent amplitude and incoherent terms beyond the WPOA phase-only model. The paper further notes that tcDPC amplitude contrast vanishes at focus and grows with $\chi$4, while tcBF amplitude contrast is maximal near $\chi$5; single-defocus datasets therefore do not uniquely separate phase and amplitude for strong scatterers or thicker specimens (Ma et al., 28 Jul 2025).

Under those constraints, tcDPC is best understood as the antisymmetric, imaginary component of tilt-corrected BF phase contrast in 4D-STEM: a direct, detector-pixel-resolved generalization of DPC that becomes especially powerful when combined with tcBF through per-image aberration correction, but remains independently useful whenever the primary target is robust, dose-efficient recovery of antisymmetric phase-gradient information (Ma et al., 1 Oct 2025).