Implicit Electric Field Conjugation (iEFC)

• iEFC is a method for coronagraphic dark-hole creation that minimizes empirically measured probe-difference signals, enabling robust high-contrast imaging.
• It bypasses traditional field reconstruction by using an empirically calibrated response matrix derived from deformable mirror probe differences, reducing model dependency.
• Applications span single and two-DM systems, fiber coupling, and photonic lantern nulling, achieving deep contrasts in laboratory and simulation studies.

Implicit Electric Field Conjugation (iEFC) is a data-driven focal-plane wavefront sensing and control method for coronagraphic high-contrast imaging in which the controller minimizes empirically measured probe-difference signals that are linearly related to the focal-plane electric field, rather than explicitly estimating that field from a diffraction model. In its canonical usage, iEFC was introduced for dark-hole creation and maintenance in the presence of non-common path aberrations and has since been extended to two-deformable-mirror coronagraphs, vacuum testbeds, single-mode-fiber coupling, and photonic-lantern nulling (Haffert et al., 2023, Milani et al., 2024, Gorkom et al., 2024, Liberman et al., 2024, Xin et al., 31 Mar 2025).

1. Definition and relation to classical electric-field control

iEFC belongs to the family of high-order wavefront sensing and control (HOWFSC) methods used to suppress coherent stellar leakage in coronagraphic instruments. Its defining distinction from standard Electric Field Conjugation (EFC) is that the controller acts on a directly measured proxy for the electric field, typically a vector of probe-difference images, instead of reconstructing the complex focal-plane field and then canceling it with a model-derived Jacobian (Milani et al., 2024).

In standard EFC, the control loop depends on model accuracy twice: once to construct the Jacobian that maps deformable-mirror (DM) commands to focal-plane field changes, and again to estimate the aberrated field from pairwise-probing measurements. Roman Coronagraph studies make this dependence explicit: the estimated field $\mathbf{E}_{ab}$ is inferred from probe images, and the control law is built from a model-derived $G_{EFC}$. iEFC removes that dependency by choosing the directly measurable probe-difference vector $\boldsymbol{\delta}$ as the control variable and empirically calibrating the corresponding response matrix $G_{IEFC}$ from images (Milani et al., 2024).

A comparative laboratory study placed iEFC alongside pairwise probing plus EFC and self-coherent-camera plus EFC as one of three complete dark-hole digging strategies. In that framing, pairwise probing and the self-coherent camera are sensing methods paired with EFC, whereas iEFC is a full wavefront sensing and control algorithm that minimizes the pairwise probe differential intensity $dI$ through an empirically measured Jacobian $G_{iEFC}=\partial(dI)/\partial\alpha$ (Desai et al., 2023).

The motivation for iEFC is strongest in systems where an accurate optical model is difficult to realize. The original iEFC paper explicitly targeted non-common path aberrations in complex ground-based instruments and argued that empirical calibration makes the method robust for systems in which Fresnel propagation, chromatic effects, polarization, DM registration, or other hardware-specific details are difficult to model with sufficient fidelity (Haffert et al., 2023).

2. Mathematical formulation

The original derivation begins from the pupil-plane field

$E = A(1+g)e^{i\phi},$

where $A$ is the pupil function, $g$ is the aberration function, and $\phi$ is the DM phase. After propagation through the instrument and coronagraph, represented by a linear operator $G_{EFC}$0, the focal-plane field is

$G_{EFC}$1

Under pairwise probing, the positive and negative probe images yield a differential signal that, in the small-phase regime $G_{EFC}$2, is linear in the coherent speckle field. With $G_{EFC}$3 probes, the measurements can be stacked as

$G_{EFC}$4

where $G_{EFC}$5 is the vector of probe-difference images, $G_{EFC}$6 is the probe response matrix, and $G_{EFC}$7 is the stacked real-imaginary representation of the speckle field (Haffert et al., 2023).

Classical EFC first reconstructs $G_{EFC}$8 by a regularized inverse,

$G_{EFC}$9

and then solves for modal DM coefficients $\boldsymbol{\delta}$0 through

$\boldsymbol{\delta}$1

iEFC collapses these two stages into a single response matrix

$\boldsymbol{\delta}$2

so that the control law becomes

$\boldsymbol{\delta}$3

This is the sense in which the electric field is handled implicitly: the controller minimizes a measurable signal that is linearly related to the field, without explicitly reconstructing the field itself (Haffert et al., 2023).

Roman Coronagraph work reformulated the same idea in a modal-response language convenient for one- and two-DM systems. With two DMs, the image-plane field is written as

$\boldsymbol{\delta}$4

where $\boldsymbol{\delta}$5. In iEFC, each calibration column is measured from images rather than computed from Fourier propagation. For calibration mode $\boldsymbol{\delta}$6,

$\boldsymbol{\delta}$7

and the control objective is

$\boldsymbol{\delta}$8

The controller therefore minimizes the measured probe-difference vector $\boldsymbol{\delta}$9, not an explicitly estimated $G_{IEFC}$0 (Milani et al., 2024).

Roman studies also standardized the contrast metric as normalized intensity,

$G_{IEFC}$1

where $G_{IEFC}$2 is the unocculted on-axis PSF image. In that literature, “contrast” denotes this normalized intensity (Milani et al., 2024).

3. Empirical calibration, probes, and control workflow

The practical core of iEFC is empirical calibration of a response matrix from camera data. In the low-jitter calibration procedure of the original paper, one applies positive and negative pokes of each DM control mode and forms a double difference,

$G_{IEFC}$3

leading to

$G_{IEFC}$4

For stronger jitter, the same paper proposed calibrating from many random linear combinations of modes,

$G_{IEFC}$5

so that the response matrix is learned from measured input-output data rather than a model (Haffert et al., 2023).

Across subsequent implementations, the mode basis and probe family are instrument-dependent. Reported calibration sets include Fourier modes, Hadamard modes, individual actuator modes, and truncated Hadamard bases; reported probes include single-actuator probes, Fourier probes, shifted Fourier probes, and power-law probes $G_{IEFC}$6 with $G_{IEFC}$7 (Milani et al., 2024, Gorkom et al., 2024, Milani et al., 2 Sep 2025). A two-DM Roman study compared four calibration sets at 825 nm for SPC-WFOV and found that Hadamard modes gave the best monochromatic result, $G_{IEFC}$8, while restricting Fourier modes only to the dark-hole spatial frequencies was suboptimal (Milani et al., 2024).

A common misconception is that “model-free” means “probe-free.” SCoOB explicitly states that iEFC still uses pairwise probes: it is “a model-free variant of EFC” that “uses double-difference probe images to construct an empirical Jacobian without a direct estimate of the electric field and attempts to minimize intensity within the targeted region” (Gorkom et al., 2024). The calibration cost can be substantial. In the JPL comparative study, iEFC required at least two probes, hence eight images per mode to build the Jacobian and four images per iteration for each subband (Desai et al., 2023). On SCoOB, typical practical refinements included truncated Tikhonov regularization, an initial oversized control region, a 10% leaky integrator, later shrinkage to the final dark hole, residual-intensity weighting for persistent speckles, and occasional Jacobian remeasurement (Gorkom et al., 2024).

4. Coronagraphic performance on laboratory testbeds and Roman simulations

The foundational iEFC paper reported that numerical experiments achieved deep contrast below $G_{IEFC}$9 with several coronagraphs and that bandwidths up to 40% could be handled without problems. On the MagAO-X internal source, the same work reported a contrast gain of a factor 10 in broadband light and a factor 20 to 200 in narrowband light, with a contrast of $dI$0 achieved with the Phase Apodized Pupil Lyot Coronagraph at $dI$1 (Haffert et al., 2023).

A comparative laboratory study at JPL tested iEFC, pairwise probing plus EFC, and self-coherent-camera plus EFC with both a vector vortex coronagraph and a scalar vortex coronagraph under matched conditions. The study found that model-free dark hole digging methods achieved comparable broadband contrasts to model-based methods. For iEFC, the reported average dark-hole contrasts were $dI$2 for the VVC at $dI$3, $dI$4 for the VVC at $dI$5, $dI$6 for the SVC at $dI$7, and $dI$8 for the SVC at $dI$9 (Desai et al., 2023).

Roman Coronagraph simulations extended iEFC to two-DM annular dark holes in the SPC-WFOV mode. In monochromatic simulation at 825 nm, the best result came from Hadamard modes at $G_{iEFC}=\partial(dI)/\partial\alpha$0, and with an unmodeled shaped-pupil-mask shear present during calibration and control the final contrast degraded only slightly to $G_{iEFC}=\partial(dI)/\partial\alpha$1. Broadband results with Hadamard modes were $G_{iEFC}=\partial(dI)/\partial\alpha$2 for a 3.6% bandpass centered at 825 nm and $G_{iEFC}=\partial(dI)/\partial\alpha$3 for a 10% bandpass; with realistic calibration noise included, those values became $G_{iEFC}=\partial(dI)/\partial\alpha$4 and $G_{iEFC}=\partial(dI)/\partial\alpha$5, respectively. The same study estimated about 6.8 hours to calibrate one noisy broadband Jacobian for 4096 Hadamard modes on $G_{iEFC}=\partial(dI)/\partial\alpha$6 Puppis (Milani et al., 2024).

SCoOB provided an experimental vacuum benchmark. In a half-sided D-shaped dark hole, the best reported mean normalized intensity was $G_{iEFC}=\partial(dI)/\partial\alpha$7 in a $G_{iEFC}=\partial(dI)/\partial\alpha$8 bandwidth, $G_{iEFC}=\partial(dI)/\partial\alpha$9 in a 2% bandwidth, and $E = A(1+g)e^{i\phi},$0 in a 15% bandwidth, with vacuum $E = A(1+g)e^{i\phi},$1 Torr and measured science-camera jitter $E = A(1+g)e^{i\phi},$2 RMS (Gorkom et al., 2024). A later SCoOB comparison paper used iEFC as the benchmark against a self-coherent camera and reported a final mean contrast of $E = A(1+g)e^{i\phi},$3 in a $E = A(1+g)e^{i\phi},$4 half-annulus dark hole, versus $E = A(1+g)e^{i\phi},$5 for the SCC implementation on the same bench (Derby et al., 2 Sep 2025).

5. Extensions beyond image-plane dark holes

Although iEFC originated as a dark-hole controller for coronagraphic images, later work generalized the method to measurement spaces that are not pixel-wise focal-plane fields. A precursor to this perspective appeared in fiber-based EFC for a single-mode optical fiber, where the controlled quantity was not a dark-hole image but the complex overlap integral of the stellar field with the fiber mode,

$E = A(1+g)e^{i\phi},$6

That study reported an initial normalized intensity of $E = A(1+g)e^{i\phi},$7 in the fiber at $E = A(1+g)e^{i\phi},$8, a final monochromatic normalized intensity of $E = A(1+g)e^{i\phi},$9, and $A$0 in $A$1 broadband light, and it explicitly framed the controller around the overlap quantity rather than image intensity (Sayson et al., 2019).

The direct iEFC extension to single-mode-fiber control formalized the same idea in model-independent form. Instead of minimizing dark-hole intensity, the controller minimizes residual starlight coupling into the fiber: $A$2 Broadband simulations showed that both fiber-based EFC and iEFC achieved a normalized intensity greater than $A$3 in the low-wavefront-error regime, but iEFC outperformed EFC by approximately 100x in the high-wavefront-error regime at 150 nm RMS WFE (Liberman et al., 2024).

A further extension applied iEFC to the Photonic Lantern Nuller, where the controlled observables are the couplings into nulled lantern ports rather than camera pixels. In that setting the controller solves

$A$4

When all four nulled ports were targeted simultaneously, null depths improved to the few-$A$5 range for several ports; in a single-port run targeting $A$6, the measured stellar coupling reached $A$7 with null depth $A$8 (Xin et al., 31 Mar 2025). These extensions show that the defining object in iEFC is not a specific image-plane geometry but an empirically calibrated linear measurement space in which stellar leakage can be minimized.

6. Robustness, operational limits, and terminological boundaries

The principal strength of iEFC is robustness to model mismatch, but that robustness is conditional rather than absolute. Roman studies showed that static shaped-pupil-mask shear degraded monochromatic two-DM iEFC only slightly, from $A$9 to $g$0, whereas model-based EFC under the same perturbation required alternating regularization, nonlinear Jacobian recomputation every 15 iterations, and 60 iterations total (Milani et al., 2024). This motivation aligns with earlier robustness studies of conventional EFC, which found strong sensitivity to unresponsive DM actuators, Lyot-stop misalignment, and focal-plane-mask misalignment in Project 1640 simulations (Matthews et al., 2017).

At the same time, empirical calibration does not eliminate alignment tolerances. A MagAO-X tolerancing study analyzed iEFC after post-calibration drift and found that focal-plane-mask offsets remained convergent up to about $g$1, Lyot-stop shifts caused noticeable degradation only near $g$2, DM misregistration began to diverge beyond about 0.5 actuator, and a pre-apodizer bump-mask offset larger than about $g$3 caused rapid divergence (Liberman et al., 2024). This suggests that iEFC is robust to some optical-model errors but only within a finite alignment envelope.

The dominant operational cost is calibration time and calibration noise. Roman broadband studies estimated about 6.8 hours for one noisy 4096-mode calibration and about 20.4 hours if three Roman narrowband HOWFSC filters were used (Milani et al., 2024). SCoOB experiments similarly showed that empirical recalibration is not negligible operationally: one reported sequence contained a gap from about 3 to 21 minutes due to the duration of the second iEFC calibration step (Gorkom et al., 2024). A practical implication is that iEFC is especially attractive when the calibrated response matrix remains valid for many control iterations.

Integration with other control loops is possible but requires explicit decoupling. On SCoOB, iEFC was combined with a Lyot-based low-order wavefront sensing and control loop using a reference-offset model so that the low-order loop would not remove the dark-hole command. In that combined architecture, baseline iEFC reached $g$4, the best combined LLOWFSC+iEFC result was $g$5, and the average maintained contrast over 120 minutes was $g$6, while the abstract summarized the result as maintenance of $g$7 contrast levels in air (Milani et al., 2 Sep 2025).

The term “iEFC” also has a narrower, non-canonical appearance outside coronagraphy. A molecular-electronics study of an anthraquinone single-molecule transistor is described in the supplied literature as highly relevant “in the language of ‘Implicit Electric Field Conjugation’,” but the mechanism there is not a focal-plane control algorithm. Rather, the gate field acts primarily by changing the molecular charge state, and that reduction changes conjugation and lifts destructive interference of transport pathways (Koole et al., 2015). In established arXiv usage, however, iEFC overwhelmingly denotes the empirical, data-driven wavefront-control method developed for coronagraphic dark-hole creation and related modal-leakage control problems (Haffert et al., 2023).