HRTEM Contrast Transfer Function

• HRTEM CTF is a mathematical function that describes the frequency-dependent transfer of spatial information in high-resolution TEM based on aberrations and imaging conditions.
• It encodes electron-optical aberrations and partial coherence effects through a sine (or cosine with phase plates) modulation, directly affecting image contrast and resolution limits.
• Advanced estimation methods like multitaper power spectral analysis and phase plate optimizations improve CTF calibration, benefiting both experimental imaging and computational reconstruction workflows.

The Contrast Transfer Function (CTF) in high-resolution transmission electron microscopy (HRTEM) encodes the frequency-dependent modulation of image contrast arising from the electron-optical aberrations and imaging conditions of the microscope. The CTF fundamentally determines which spatial frequencies of the sample are transferred to the recorded image, and with what sign, magnitude, and fidelity. Its precise mathematical structure and dependence on imaging parameters critically impact image interpretation, resolution limits, CTF correction protocols, and, more recently, the assessment of information content for computational analysis workflows.

1. Mathematical Foundations of the HRTEM Contrast Transfer Function

The HRTEM CTF is derived by considering the imaging of a weak-phase specimen of projected electrostatic potential $V(r)$, illuminated by a plane electron wave. The exit-wave $\psi_f(q)$ in reciprocal space after lens propagation is expressed as

$$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$

where $\sigma$ is the interaction constant, $\mathcal{F}$ the Fourier transform, and $H(q)$ the complex lens transfer function (DaCosta et al., 9 Dec 2025). $H(q)$ factorizes into the aperture function $A(q)$, the envelope $E(q)$ (accounting for partial coherence and aberration damping), and the aberration-induced phase term $e^{-i\chi(q)}$:

$\psi_f(q)$0

Under the weak-phase approximation, the image intensity’s linear component is governed by the “phase-contrast transfer function”:

$\psi_f(q)$1

The phase shift $\psi_f(q)$2 is typically expanded as

$\psi_f(q)$3

with $\psi_f(q)$4 the electron wavelength, $\psi_f(q)$5 defocus, $\psi_f(q)$6 the spherical aberration coefficient, and $\psi_f(q)$7 the spatial frequency (Gamm et al., 2010, Heimowitz et al., 2019). Further generalizations include higher-order aberrations and astigmatism via tensorial polynomial expansions (DaCosta et al., 9 Dec 2025).

2. Physical Interpretation and Role in Image Formation

$\psi_f(q)$8 determines the amplitude and sign with which spatial frequencies of the object are transferred to the image. Its oscillatory structure stems from the sinusoidal dependence on $\psi_f(q)$9, resulting in alternating frequency bands with positive, negative, or zero transmission:

• At spatial frequencies for which $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$0, $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$1, the CTF crosses zero and information is lost (so-called Thon rings).
• The first CTF zero determines the effective HRTEM resolution under given defocus and aberration conditions (“Scherzer resolution”).

Partial coherence (temporal and spatial) damps $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$2 via envelope functions:

$$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$3

where $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$4 is the defocus spread, $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$5 the illumination semi-angle. The combined envelope $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$6 ultimately limits the information accessible in the image (Gamm et al., 2010).

3. Phase Plate Modification and Optimization Regimes

The canonical CTF has the sine form, $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$7, resulting in minimal transfer near zero frequency. Introduction of a physical phase plate in the objective back focal plane that imparts a phase shift $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$8 between unscattered and scattered electrons transforms the CTF to $$\psi_f(q) = \mathcal{F}_{r\to q}\left[\psi_0(r) \exp(i \sigma V(r))\right] H(q)$$9:

$\sigma$0

This conversion shifts all low-frequency CTF zeros to maxima, drastically improving low- and mid-frequency contrast and flattening the CTF response (Gamm et al., 2010). Under idealized conditions ($\sigma$1, $\sigma$2, $\sigma$3), the CTF becomes flat up to the information limit set by $\sigma$4, enabling maximal, non-oscillatory contrast transfer across all spatial frequencies within the coherence envelope.

A summary of the effect of phase plates is presented below:

CTF Type	Mathematical Form	Low-Frequency Behavior	Practical Benefits
Without Phase Plate	$\sigma$5	Zero at $\sigma$6	Need for specific focus (“Scherzer”)
$\sigma$7 Physical Phase Plate	$\sigma$8	Maximum at $\sigma$9	Strong low-/mid-frequency contrast, relaxed focus tolerances

The improvement is pronounced in aberration-corrected instruments, enabling near “in-focus” operation with broadened tolerances: for 200 keV systems, $\mathcal{F}$0m and $\mathcal{F}$1 nm yield <5% contrast loss, much more forgiving than traditional conditions (Gamm et al., 2010).

4. Estimation, Fitting, and Zero-Crossings of the CTF

Experimental images require precise CTF calibration since $\mathcal{F}$2 acts as a band-pass filter with sign inversions and amplitude modulation. The estimation process proceeds as follows (Heimowitz et al., 2019):

• The modulus of the Fourier transform of the image is analyzed for CTF-induced oscillations (Thon rings), serving as fiducials for the determination of $\mathcal{F}$3’s parameters (defocus, astigmatism, $\mathcal{F}$4).
• Advanced estimation schemes, such as the multitaper power spectral density estimator, mitigate bias/variance by exploiting orthonormal tapers (DPSS windows) and averaging over image blocks.
• Background subtraction (via convex programming over angularly averaged spectra) and steerable-basis denoising enable isolation of CTF features even under low-SNR conditions.

Zero-crossings of $\mathcal{F}$5 are identified as minima in the denoised, background-subtracted spectra; their loci (rings in reciprocal space) are fitted to the expected forms via nonlinear optimization, solving for $\mathcal{F}$6 (defocus, astigmatism axes). Accurate zero-crossing localization is critical, as it ensures correct restoration and sign assignment in frequency bands during phase-flipping and Wiener deconvolution (Heimowitz et al., 2019).

5. CTF and Quantification of Information Content

$\mathcal{F}$7 directly quantifies the information content available at each frequency. For analyses of experimental robustness and computational generalization, scalar information metrics based on $\mathcal{F}$8 have been developed (DaCosta et al., 9 Dec 2025):

• Total-information fraction $\mathcal{F}$9:

$H(q)$0

This gives the squared proportion of ideal envelope-limited information passed by the CTF under given aberration conditions.

• Pass-band overlap $H(q)$1:

$H(q)$2

Comparing two sets of imaging conditions, $H(q)$3 measures the overlap of their pass-bands. Values near unity indicate similar information transfer; lower values signal mismatch in frequency response.

These metrics mechanistically predict the stability of downstream analyses (e.g., neural network segmentation) under distributional shifts in imaging parameters. Performance degrades smoothly as $H(q)$4 and $H(q)$5 decrease; highest generalization is observed when training and testing CTFs have aligned pass-bands and comparable information content (DaCosta et al., 9 Dec 2025).

6. Impact on Experimental and Computational Workflows

Accurate knowledge and control of the HRTEM CTF impact both experimental acquisition and computational reconstruction. Experimentally, tuning imaging parameters to optimize $H(q)$6 (such as by adopting phase plates and aberration correction strategies) can deliver strong, localized contrast with minimized delocalization and broadened tolerances (Gamm et al., 2010). Computationally, precise CTF estimation:

• Enables phase-flipping and Wiener filtering to maximize retrieved structural information.
• Enhances signal-to-noise in feature extraction and particle picking for single-particle analysis.
• Provides principled metrics for evaluating the out-of-distribution robustness of machine learning models, as shown by the use of CTF-based $H(q)$7 and $H(q)$8 in large-scale neural network performance studies (DaCosta et al., 9 Dec 2025).

State-of-the-art multitaper estimation combined with steerable-basis denoising and robust zero-crossing fitting yields consensus-class defocus and astigmatism parameters, supporting real-time feedback and motion correction pipelines (Heimowitz et al., 2019).

7. Practical Recommendations and Frontier Challenges

Guidelines for ideal imaging include minimizing residual $H(q)$9 and defocus via aberration correction, using stable phase plates for flat CTFs, and explicitly measuring coherence envelopes to set the information limit. Correction of higher-order aberrations is necessary when pushing resolution beyond 0.08 nm (Gamm et al., 2010). In computational settings, reporting CTF specification and leveraging CTF-aware information metrics is essential for reproducible performance analysis (DaCosta et al., 9 Dec 2025).

Continued development of robust, low-variance CTF estimators, particularly under severe noise and heterogeneity, and further integration of CTF theory into machine learning assessment frameworks remain active areas of research. A plausible implication is that the classical theory, encapsulated in $H(q)$0, $H(q)$1, and $H(q)$2, will increasingly serve as a physical benchmark for both experimental optimization and computational generalization in contemporary and future HRTEM applications.