MZI Mirror: Functions & Implementations

• MZI mirrors are optical elements or engineered equivalents that define, fold, and modulate interferometric paths by controlling phase differences.
• In free-space designs, mirrors enable precise beam routing and stable interference, evidenced by metrics like sub-degree Allan deviation and near-unity visibility.
• Alternative implementations leverage Bragg reflectors, quantum emitters, and integrated designs to achieve effective reflection while reducing noise and expanding bandwidth.

Searching arXiv for recent and foundational papers on Mach–Zehnder interferometer mirrors and mirror-related architectures. A Mach–Zehnder interferometer (MZI) mirror is not a single standardized component but a class of reflector-related elements that define, fold, or modulate the two interfering paths of an MZI. In free-space realizations, mirrors route the beams and determine the geometric accessibility and common-mode stability of the arms; in plasmonic, waveguide-QED, and optomechanical realizations, mirror-like behavior can instead be implemented by Bragg reflectors, beam-splitter analogs, or quantum scatterers. Across these implementations, the physically relevant role is the same: the mirror or mirror-equivalent element contributes to path formation, phase accumulation, and recombination sensitivity, while the overall interferometric response remains governed by the relative phase between the two arms (Micuda et al., 2014).

1. Mirror function within Mach–Zehnder interferometry

A conventional MZI is most compactly understood as a two-path interferometer in which an input field is split, acquires a relative phase difference, and is recombined. In that sense, the essential observable is not a mirror in isolation but the phase-dependent routing of amplitude between the outputs. In a quantum-optical formulation, an MZI is described as two beam splitters plus two mirrors, with the path-length difference encoded as a phase delay $\varphi=\omega\tau$ (Ataman, 2014). For a single photon entering one input port, the output probabilities depend on the accumulated phase, e.g. in the single-MZI treatment $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$ and $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$, which captures the interferometer’s phase-sensitive routing behavior (Ataman, 2014).

This formal point matters because the term “MZI mirror” is often used loosely. In some architectures, the mirrors are literal reflectors that fold the arms; in others, there are no discrete mirrors at all. A fiber MZI built from two 50:50 couplers explicitly replaces the “mirror-like” arm termination of folded geometries by the second coupler and the fiber path geometry (Xavier et al., 2011). Likewise, a silicon ITO MZI modulator uses Y-junction splitters and recombiners and explicitly does not introduce mirrors, reflectors, cavity mirrors, photonic-crystal mirrors, or mirror-based feedback paths (Gui et al., 2021). This establishes a useful boundary condition: an MZI mirror is implementation-dependent rather than topologically mandatory.

A recurring misconception is therefore that an MZI always contains mirrors in the same sense as a Michelson interferometer. The literature in the provided corpus shows the opposite. Free-space and folded MZIs often rely on mirrors for path routing, whereas integrated traveling-wave MZIs can be entirely mirrorless (Gui et al., 2021). The relevant invariant is the two-path interference condition, not the specific reflector technology.

2. Free-space mirrors: path folding, accessibility, and passive stability

The most explicit free-space treatment is the displaced-Sagnac Mach–Zehnder interferometer reported as an MZDS interferometer (Micuda et al., 2014). It is realized with a single 1″ beam splitter cube and three mirrors. The beam splitter both splits and recombines the light, while the mirrors fold the two arms into a displaced Sagnac configuration with 8 mm spatial separation and 1.34 m arm length. The design goal is to recover much of the common-path stability of a Sagnac interferometer while preserving individually accessible arms, which are not available in a true common-path Sagnac geometry (Micuda et al., 2014).

In this device, mirror placement is inseparable from system-level performance. The 8 mm separation is described as a compromise between the clear aperture of the optical components and the need for convenient individual addressing of the beams. After alignment, all three mirrors are kept fixed; interference is not scanned with a piezo-mounted mirror, because that is stated to be inconvenient in the displaced Sagnac geometry. Instead, relative phase is adjusted by tilting one of the glass plates (Micuda et al., 2014). This fixes the mirrors as static geometric elements rather than dynamic actuators.

The quantitative performance of that mirror-defined free-space geometry is notable. The interferometer has a footprint of 27 × 40 cm, shows phase deviation less than 0.4 deg during a 250 s measurement, and has Allan deviation below 3 deg or 7 nm for 1.5 hours without active stabilization (Micuda et al., 2014). The minimum Allan deviation is reported as less than 0.39 deg at an optimal integration time of about 250 s, corresponding to an arm-length deviation of 0.87 nm and a relative length deviation of $0.65\times 10^{-9}$ (Micuda et al., 2014). The same work states that the closeness of the two paths makes the interferometer virtually immune to acoustic waves up to tens of kHz, which directly links mirror geometry to passive environmental rejection (Micuda et al., 2014).

Polarization behavior further clarifies the role of the mirrors. The measured reflectance of the mirrors shows only negligible dependence on polarization. Although the mirrors induce a small phase shift between S and P polarizations, this does not influence interferometer performance because both arms experience the same phase shift. When stricter polarization fidelity is required, low-dephasing dielectric mirrors are recommended; OA019 from Femto-Optics is specifically mentioned, with measured S–P phase shift smaller than 1 deg (Micuda et al., 2014). In this geometry, mirror-induced phase is therefore largely common-mode rather than differential.

3. Mirror-induced phase, visibility, and polarization fidelity

In mirror-bearing MZIs, the practical question is not whether a reflection produces phase, but whether that phase appears symmetrically in both arms. The displaced-Sagnac free-space implementation makes this distinction explicit: the small S–P mirror phase shift does not degrade visibility because both arms traverse the same mirror-induced polarization phase structure (Micuda et al., 2014). The dominant residual polarization sensitivity is instead attributed to the beam splitter splitting ratio, measured as 45:55 for horizontal polarization and 43:57 for vertical polarization (Micuda et al., 2014).

The measured visibilities in that system separate mirror effects from mode-matching and splitter imbalance. With single-mode-fiber-coupled detection, the visibility at D1 is $99.96 \pm 0.01\%$ for H polarization and $99.98 \pm 0.01\%$ for V polarization, while at D2 it is $97.33 \pm 0.03\%$ for H and $93.60 \pm 0.09\%$ for V (Micuda et al., 2014). Bulk photodiode measurements yield lower values, and the authors conclude that single-mode collection improves mode matching and increases visibility by about 2% relative to direct bulk detection (Micuda et al., 2014). The approximately 3.5% H/V discrepancy in one output port is attributed mainly to the beam splitter rather than to the mirrors (Micuda et al., 2014).

The visibility definition used there is Michelson’s formula, written in the paper as the standard expression in terms of maximum and minimum intensities at a given output port (Micuda et al., 2014). The same visibility logic recurs in other MZI variants. In the plasmonic surface-polariton MZI based on an elliptical Bragg mirror, the complementary output fringes are modeled through beam-splitter reflection and transmission coefficients, and the visibility is $V=(I_{\max}-I_{\min})/(I_{\max}+I_{\min})$; with $|R|^2\simeq|T|^2\simeq 1/2$, the visibility approaches unity (Drezet et al., 2010). In fiber-based long-arm operation, active stabilization enables a single-photon net visibility of $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$0 in a 1 km interferometer (Xavier et al., 2011). These results collectively suggest that when mirrors are used as passive path-folding elements, high visibility depends more critically on symmetry, mode overlap, and splitter balance than on reflection itself.

A related misconception is that any mirror in an MZI is automatically a dominant source of polarization or phase error. The free-space displaced-Sagnac study does not support that claim: mirror reflectance is nearly polarization independent, and the main H/V visibility imbalance is assigned to the beam splitter (Micuda et al., 2014). This suggests that in carefully symmetric layouts, mirror-induced phase shifts are often common-mode nuisances rather than first-order limitations.

4. Mirror equivalents: Bragg reflectors, quantum mirrors, and beam-splitter analogs

Outside conventional free-space optics, mirror functionality in an MZI is frequently realized by structures that are reflective only in an effective or mode-dependent sense. A particularly clear example is the surface plasmon polariton MZI implemented on a gold film with an elliptical Bragg mirror (Drezet et al., 2010). There, SPPs launched at one focus of the ellipse are reflected by a Bragg mirror formed from five confocal ellipses of gold protrusions and directed toward the second focus, where a protrusion-based beam splitter divides the SPP into two outputs. The Bragg condition is chosen with $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$1 and $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$2 for $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$3, yielding mirror reflectivity up to 90% (Drezet et al., 2010). In that architecture, the elliptical Bragg mirror simultaneously provides directional reflection and arm formation.

A more radical reinterpretation appears in waveguide QED, where a single two-level emitter in a 1D waveguide can act as a quantum mirror or a quantum beamsplitter (Almeida et al., 2019). In the monochromatic resonant limit, the reflected probability becomes $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$4 and the transmitted probability $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$5, so the emitter functions as a quantum mirror (Almeida et al., 2019). Off resonance or at finite linewidth, the same element can realize balanced splitting with $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$6 under the conditions identified in the paper (Almeida et al., 2019). When two such emitters are concatenated, they form a fully quantum Mach–Zehnder interferometer whose behavior can mimic a classical MZI, an MZI with a $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$7 phase shift, or an MZI with a $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$8 phase shift, depending on detuning and linewidth (Almeida et al., 2019).

Frequency-domain interferometry provides yet another mirror-equivalent picture. A frequency-domain MZI implemented with two PPLN waveguides uses partial quantum frequency conversion as the analog of beam splitting between two frequency modes, 1580 nm and 795 nm (Kobayashi et al., 2017). Here the “arms” are frequencies, not spatial paths. The nonlinear converters act as frequency-domain beam splitters, and the second converter recombines the amplitudes. At about 50% internal conversion efficiency, visibilities of 0.99 are obtained at both outputs (Kobayashi et al., 2017). Although this is not a mirror architecture in the conventional sense, it makes clear that what an MZI requires is coherent amplitude redirection and recombination, not necessarily geometric reflection.

These examples show that “mirror” in MZI research can denote a literal reflector, a Bragg-reflecting nanostructure, or a resonant scatterer whose interference with re-emission produces effective reflection. The common physical content is coherent path definition.

5. Mirrors as active elements: micromirrors, delay mirrors, and path control

In some MZI realizations, the mirror is not merely a passive routing element but the locus of the signal transduction itself. A quantum optomechanical MZI replaces one of the two fixed mirrors with an oscillating quantum micromirror in one arm, while the other arm contains a fixed mirror plus phase shifter (Barchielli et al., 2021). The lower-arm field interacts with the mirror through radiation-pressure scattering, and the outgoing field satisfies $P_4=\frac{1}{2}(1+\cos(\omega\tau_1))$9, so the mirror acts as a phase modulator whose phase depends on the mechanical position operator $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$0 (Barchielli et al., 2021). In that configuration, no cavity is involved; the interaction is explicitly with traveling Bose fields. The output difference current after the second beam splitter is interpreted as an effective homodyne measurement, and in the strong-laser, weak-coupling regime the generated light can show a negative Mandel $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$1-parameter and nearly complete cancellation of shot noise in the difference-current spectrum (Barchielli et al., 2021).

Delay interferometers provide a different active use of mirrors. In a compensation-free free-space delay interferometer for phase-encoded QKD, the long-arm mirror is translated to tune the delay, and the delay itself is created simply by mirror displacement rather than by refractive compensation elements (Tretiakov et al., 10 Apr 2025). The output field is the coherent sum of the two arm fields, and the interference visibility is determined by the normalized field-overlap integral between the propagated transverse profiles (Tretiakov et al., 10 Apr 2025). Measured visibility values up to about 0.95 are reported for a delay of 2 ns, with the beam radius playing the central role in preserving overlap (Tretiakov et al., 10 Apr 2025). Here the mirror is the delay actuator, but the limiting physics is spatial multimode overlap.

Active stabilization in long-fiber MZIs uses a different kind of path-control element. A 1 km fiber MZI employs a piezoelectric fiber stretcher for phase stabilization and an electro-optic phase modulator for continuous user-controlled phase adjustment, achieving stable single-photon interference with net visibility 0.97 (Xavier et al., 2011). Although no discrete mirrors are present, this case clarifies by contrast what mirrors do in free-space MZIs: they provide the geometrical degree of freedom that fiber systems replace with controlled optical path length in waveguide media (Xavier et al., 2011).

Taken together, these studies show that mirrors in MZI systems can be passive reflectors, quantum transducers, or delay-setting elements. The unifying requirement is that they modify the differential optical phase in a controllable way without erasing coherence.

6. Mirror minimization, mirror replacement, and architectural tradeoffs

A substantial part of the MZI literature is motivated not by better mirrors, but by reducing dependence on problematic ones. In pass-through Mach–Zehnder/Fabry–Pérot topologies for macroscopic quantum measurements, the main advantage relative to Michelson/Fabry–Pérot interferometers is precisely that the design does not contain high-reflectivity end mirrors with multilayer coatings (Khalili, 2011). The arm cavities use equal moderate transmittance on both mirrors rather than an extremely reflective end test mass. The paper gives a concrete comparison: a Michelson/Fabry–Pérot prototype example with $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$2 and $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$3 corresponds to about 23 coating layers total per cavity, whereas the Mach–Zehnder/Fabry–Pérot case with $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$4 corresponds to about 12 layers total (Khalili, 2011). The stated consequence is roughly a factor of two reduction in coating thermal noise spectral density (Khalili, 2011).

This architectural perspective is important because it reverses the usual intuition that adding mirrors improves interferometric control. In precision metrology, high-reflectivity mirrors can dominate the noise budget through coating Brownian noise, so an MZI topology may be preferred specifically because it avoids such mirrors (Khalili, 2011). Conversely, in free-space displaced-Sagnac operation, three fixed mirrors are central to achieving passive long-term stability and individually accessible arms (Micuda et al., 2014). The preferred mirror strategy is therefore application-specific.

The integrated photonics literature reinforces this point from the opposite side. The ITO-based silicon MZI modulator explicitly avoids cavity and mirror architectures because resonators and mirror sections can provide high extinction ratio but at the cost of reduced bandwidth and additional footprint (Gui et al., 2021). The device remains a standard two-arm traveling-wave interferometer with no discrete mirrors (Gui et al., 2021). Likewise, narrowband integrated MZIs based on anti-symmetric multimode waveguide Bragg gratings use grating-assisted mode-selective reflection and asymmetric Y-branches to create an equivalent narrowband coupler only inside a designed wavelength band, while remaining transparent outside that band (Wang et al., 29 May 2026). Such systems reproduce selected mirror-like spectral functions without reverting to conventional mirror-based free-space layouts.

The broader implication is that an “MZI mirror” is best treated as a functional category. Depending on the platform, the design objective may be to exploit mirrors for stability and access, to turn a scatterer into a quantum mirror, or to avoid high-reflectivity mirrors altogether.

7. Representative implementations and research directions

The diversity of mirror roles across MZI research can be summarized concisely.

Implementation	Mirror or mirror-equivalent role	Representative result
Displaced-Sagnac free-space MZI	Three mirrors fold two 1.34 m arms with 8 mm separation	Allan deviation less than 0.39 deg; visibility above 93% (Micuda et al., 2014)
SPP MZI	Elliptical Bragg mirror forms and redirects SPP arms	Reflectivity up to 90%; unit-visibility fringes (Drezet et al., 2010)
Quantum waveguide-QED MZI	Two-level emitter acts as quantum mirror or quantum beamsplitter	On resonance in monochromatic limit, $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$5 (Almeida et al., 2019)
Optomechanical MZI	Oscillating micromirror imprints quantum phase by radiation pressure	Negative Mandel $P_5=\frac{1}{2}(1-\cos(\omega\tau_1))$6 and nearly complete shot-noise cancellation in difference current (Barchielli et al., 2021)
Pass-through MZI/FP	Topology avoids high-reflectivity end mirrors	About 12 coating layers versus 23 in the cited Michelson/FP comparison (Khalili, 2011)
Traveling-wave integrated MZI	No discrete mirrors	Mirror/cavity architectures are explicitly avoided (Gui et al., 2021)

This spread of implementations suggests two stable conclusions. First, the mirror in an MZI is not a universal part but a platform-specific solution to path definition, phase control, or reflection engineering. Second, the most technically consequential questions are rarely about reflectivity alone; they concern whether the mirror-induced phase is common-mode or differential, whether the mirror preserves polarization and spatial overlap, and whether replacing or removing mirrors improves the relevant stability-noise-bandwidth tradeoff.

In quantum-information and sensing contexts, the displaced-Sagnac free-space design shows why carefully chosen mirrors remain attractive when individually accessible arms, low polarization sensitivity, no active phase lock, and compatibility with single-photon detection are required (Micuda et al., 2014). In nanophotonics and waveguide QED, Bragg reflectors and TLS scatterers show that mirror behavior can be synthesized rather than inherited from bulk optics [(Drezet et al., 2010); (Almeida et al., 2019)]. In precision measurement, pass-through MZI topologies demonstrate that the most valuable mirror may be the one eliminated from the design (Khalili, 2011).