Metastable CeZrO_4−δ polymorph for high-sensitivity nitrogen dioxide detection

Abstract

Accurate, real-time detection of nitrogen dioxide (NO₂) is essential for enforcing next-generation vehicle emission standards and ensuring environmental safety. Electrochemical devices offer a cost-effective solution; however, current sensing materials have limited sensitivity, stability, and selectivity. Here, we report a metastable polymorph of CeZrO_4−δ, stabilized by liquid-nitrogen quenching, as an NO₂-sensing material. The metastable tetragonal structure features surface-enriched Ce³⁺ species, oxygen-vacancy-related environments, and distorted tetragonal surface configurations, which collectively facilitate NO₂ adsorption and charge transport. The resulting sensor achieves a current response of 24,000% to 10 ppm NO₂, defined as [|(I_gas − I_air)/I_air|] × 100%, compared to 3300% for the equilibrium cubic phase, while maintaining stable operation under variable humidity and temperature. These findings demonstrate that metastable phases offer an effective route to tune sensing materials, and the current example provides a platform for sensitive and stable NO₂-resolved monitoring relevant to Euro 7-oriented emission diagnostics.

Reliable nitrogen dioxide sensing under changing conditions remains challenging. Here, the authors use rapid cooling to stabilize metastable cerium–zirconium oxide, enabling sensitive and stable real-time vehicle exhaust monitoring.

Introduction

The rapid phase-out of conventional fuel vehicles under the European Green Deal has elevated nitrogen dioxide (NO₂) as a critical metric for emission compliance¹. The newly enacted Euro 7 framework—Regulation (EU) 2024/1257 with Implementing Regulations (EU) 2025/1706/1707—retains light-duty NO_x limits of 60 mg·km⁻¹ for positive ignition (PI) and 80 mg·km⁻¹ for compression ignition (CI) vehicles, and requires on-board monitoring (OBM) to report tailpipe NO_x concentration and mass flow as 1-s averages updated at least once per second, enabling real-time, high-accuracy NO₂-capable sensing under dynamic driving conditions². Although Euro 7 does not prescribe a formal minimum detectable NO₂ concentration in ppm, its NO_x limits correspond to tailpipe NO₂ levels of a few to tens of ppm under representative light-duty conditions. These regulatory requirements expose the inadequacies of existing technologies, which often fail to capture transient, low-concentration variations of NO₂ amid fluctuating temperature and humidity. Consequently, the development of reliable, selective, and durable sensors has become an essential technological priority for environmental surveillance, verifying emission compliance and enabling sustainable transportation^3–5.

Optical, metal-oxide-semiconductor (MOS), and electrochemical approaches are presently the leading applicable NO₂ detection technologies^6–8. Optical techniques can deliver ppb-level precision and excellent selectivity; however, their bulky optics, high power consumption, and relatively high cost impede deployment in compact on-board platforms^9–11. MOS sensors—typically based on WO₃, SnO₂, or In₂O₃—offer simple architectures but generally require elevated operating temperatures (often > 200–300 °C) and suffer from cross-sensitivity and limited selectivity under real-world humidity and multi-gas backgrounds^12–14. Electrochemical sensors enable low-temperature operation and miniaturization¹⁵; however, they are prone to humidity-induced drift, electrolyte dehydration/flooding, and electrode aging, which together constrain long-term stability and quantitative accuracy¹⁶. In fuel-cell-type electrochemical NO₂ sensors, the gas-facing electrode must not only mediate charge transfer, but also efficiently capture trace NO₂ under humid conditions. Although Pt/C remains a conventional amperometric benchmark, recent high-performance room-temperature NO₂ sensors based on defect-engineered nonprecious materials have shown that tailored adsorption-active sites and surface redox chemistry can outperform generic noble-metal activity in NO₂ sensing^17–19. Defective Ce-Zr-O pyrochlore-related phases, especially Ce₂Zr₂O₇/Ce₂Zr₂O_7+z, provide important precedents for redox-active cerium zirconates. These Ce³⁺-containing oxygen-deficient phases have been extensively studied for oxygen intercalation and deintercalation, oxygen storage, and catalytic reactivity, because their defect chemistry and the Ce³⁺-Ce⁴⁺ redox couple strongly influence reducibility and oxygen activation^20–25. However, these studies mainly concern pyrochlore-related redox transformations under reducing and oxidizing conditions rather than room-temperature electrochemical gas sensing. Here, we address these limitations through a metastable-phase engineering strategy, in which rapid cryogenic quenching stabilizes tetragonal CeZrO_4−δ as a room-temperature electrochemical NO₂-sensing material while simultaneously regulating polymorph structure, defect chemistry, and surface states.

We demonstrate that metastable tetragonal CeZrO_4−δ, stabilized via liquid-nitrogen quenching, serves as an active sensing material for electrochemical NO₂ detection. The metastable phase, characterized by enriched Ce³⁺ species, abundant oxygen vacancies, and distorted tetragonal surface environments, promotes efficient NO₂ adsorption, rapid charge transport, and accelerated surface redox reactions. As a result, the corresponding sensor achieves a current response of 24,000% to 10 ppm NO₂, eightfold higher than that of the equilibrium cubic counterpart, at room temperature. This metastable-phase strategy provides an approach to sensitive and humidity-tolerant NO₂-resolved sensing relevant to Euro 7-oriented NO_x/OBM diagnostics.

Results and discussion

Material characterization and stabilization of metastable polymorphs

Polycrystalline CeZrO_4−δ was synthesized through a sol-gel route. Samples were preheated to 300, 600, or 900 °C (labeled CeZrO_4−δ-300, etc.) and quenched into liquid nitrogen (Supplementary Movie 1). Two control samples were obtained by slow furnace cooling (CeZrO_4−δ-S) and rapid air cooling (CeZrO_4−δ-Q). X-ray diffraction (XRD) reveals sample variations with quenching temperature (Supplementary Fig. S1), and Rietveld refinements using neutron powder diffraction (NPD) data confirm the coexistence of the stable cubic c-CeZrO_4−δ polymorph (space group Fm3̅m) and a metastable tetragonal t-CeZrO_4−δ polymorph (P4₂/nmc) (Supplementary Figs. S2–S7, Supplementary Tables S1–S14)^26,27. The t-CeZrO_4−δ fraction has a maximum value of ~ 91% at 600 °C as analyzed by XRD. Consistently, NPD analysis yields a t-CeZrO_4−δ fraction of ~ 95%, further confirming the XRD results (Fig. 1a–c). In situ variable-temperature XRD (VT-XRD) further shows that kinetic arrest at 600 °C maximizes the tetragonal phase, establishing rapid quenching as an effective route to stabilize this otherwise metastable configuration (Supplementary Fig. S8). Magnified VT-XRD patterns (Supplementary Figs. S9–S11) display distinct tetragonal reflections whose intensity maximizes at 600 °C, in agreement with XRD and NPD refinements. These results show that kinetic arrest by rapid quenching stabilizes the metastable tetragonal phase at this temperature. In this context, although the defect chemistry of the quenched samples is more appropriately described as CeZrO_4−δ, the XRD/NPD results still support coexisting cubic and tetragonal CeZrO₄-based polymorphs as the dominant average structures, rather than a dominant pyrochlore phase.

Fig. 1. Structural and electronic characterization of quenched CeZrO_4−δ polymorphs.

a Rietveld refinement of the neutron powder diffraction pattern of CeZrO_4−δ-600. The symbols, fitted profile and difference curve represent the observed, calculated and residual intensities, respectively; vertical tick marks indicate the Bragg reflection positions of the cubic and tetragonal phases. b Refined crystal structures of cubic c-CeZrO_4−δ (Fm3̅m) and tetragonal t-CeZrO_4−δ (P4₂/nmc) polymorphs. c Refined fractions of cubic c-CeZrO_4−δ and tetragonal t-CeZrO_4−δ phases. The x-axis denotes sample categories: Pristine CeZrO_4−δ, 300, 600 and 900, liquid-nitrogen-quenched samples treated at the corresponding temperatures (°C); S, slow furnace cooling; and Q, rapid air cooling. d Ce 3 d X-ray photoelectron spectroscopy (XPS) spectra of CeZrO_4−δ and CeZrO_4−δ-x (x = 300, 600 and 900). Detailed fitting parameters are provided in Supplementary Data 2. e O 1 s XPS spectra of CeZrO_4−δ and CeZrO_4−δ-600. The inset compares the fraction of chemisorbed oxygen (O_c) obtained from peak fitting for the two samples. f, g Ce L₃-edge and Zr K-edge X-ray absorption near-edge structure (XANES) spectra, respectively. The insets show enlarged views of the corresponding absorption-edge regions. The arrows indicate the lower-energy shift at the Ce L₃-edge and the higher-energy shift at the Zr K-edge. h Fourier-transform magnitudes of the k³-weighted Zr K-edge extended X-ray absorption fine structure (EXAFS) signals in R space. i Continuous wavelet-transform (WT) contour maps of the k³-weighted Zr K-edge EXAFS signals for CeZrO_4−δ (left) and CeZrO_4−δ-600 (right). Source data are provided as a Source Data file.

Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) show that quenching preserves the nanoblock morphology while simultaneously introducing surface roughness, lattice rearrangement, and concentric diffraction rings characteristic of polycrystalline materials (Supplementary Figs. S12, S13). Energy-dispersive X-ray spectroscopy (EDS) mapping verifies uniform elemental distribution (Supplementary Fig. S14), and Brunauer-Emmett-Teller (BET) analysis shows negligible variation across all samples (Supplementary Fig. S15), indicating that the sensing enhancement shown later does not arise from changes in surface area.

X-ray photoelectron spectroscopy (XPS) and X-ray absorption fine structure (XAFS) analyses were performed to probe the electronic structure of as-prepared CeZrO_4−δ and quenched samples. The high-resolution Ce 3 d spectra reveal coexisting Ce³⁺/Ce⁴⁺ states, with CeZrO_4−δ-600 showing a clear shift toward lower binding energy and the highest Ce³⁺/Ce⁴⁺ peak-area ratio (Fig. 1d, Supplementary Fig. S16–S18 and Supplementary Table S15). This shift reflects oxygen-vacancy formation at the surface, which modulates the local electronic environment and underpins improved redox activity^28,29. O 1 s spectra further show that CeZrO_4−δ-600 contains a significantly higher fraction of oxygen vacancies and chemisorbed oxygen, with the adsorbed oxygen component increasing from 1.73% to 17.55% (Fig. 1e). This enrichment of surface-active oxygen species is consistent with the increased Ce³⁺ population. Ce L₃-edge X-ray absorption near-edge structure (XANES) of CeZrO_4−δ-600 displays a subtle edge shift to lower energy, corroborating its higher Ce³⁺ population (Fig. 1f and Supplementary Fig. S19)³⁰. Raman spectra further show that CeZrO_4−δ-600 has a broader and more asymmetric fluorite-like F_2g band near 470 cm⁻¹ and a more pronounced defect-related broad feature at 600–625 cm⁻¹, while a weak band near 300–315 cm⁻¹ is generally associated with oxygen displacement from ideal fluorite positions (Supplementary Fig. S20)^31,32. These features support quenching-induced oxygen sublattice distortion and oxygen-vacancy-related disorder, consistent with the XPS, XANES, and EXAFS results. In contrast, Zr K-edge spectra remain largely unchanged, though extended X-ray absorption fine structure (EXAFS) analysis reveals shorter Zr-O bond lengths and reduced Zr-Zr coordination intensity (Fig. 1g, h). Wavelet transform (WT) analysis further shows increased local disorder after quenching (Fig. 1i and Supplementary Fig. S21), consistent with the cubic-to-tetragonal transition. Such lattice distortion plays a critical role in tuning the electronic structure, facilitating charge transport, and promoting surface reactions^33–35. These findings identify lattice distortion and defect engineering as design principles for tailoring redox-active oxides for electrochemical applications.

Gas sensing properties

Electrochemical devices based on quenched CeZrO_4−δ-600 exhibit the highest NO₂ response in the series (Fig. 2a–c and Supplementary Fig. S22–S25). At 10 ppm, the relative current change reaches 24,000%, nearly eightfold higher than that of cubic CeZrO_4−δ, with rapid response and recovery times of 7 s and 8 s, respectively. The sensor shows a linear detection range from 50 ppb to 10 ppm, with a sensitivity of 559 nA ppm⁻¹, more than three times higher than the unquenched sample (inset of Fig. 2b). The CeZrO_4−δ-600 retained 98.1% of its initial response after 98 days, indicating stable long-term operation (Fig. 2d). Stabilization of the metastable tetragonal polymorph is central to this performance gain over equilibrium phases. Benchmarking against representative NO₂ sensors shows that CeZrO₄−δ-600 combines a large response-current change with extended reported stability among the electrochemical sensors considered (Fig. 2e and Table 1). To further justify the material choice beyond the literature benchmarking in Fig. 2e, we compared CeZrO_4−δ-600 with a conventional Pt/C-based device under the same sensor configuration and test conditions (Supplementary Fig. S26). Over 0.5-20 ppm NO₂, Pt/C showed only a weak response of about 20–130 nA, whereas CeZrO_4−δ-600 increased from about 260 nA to 11,200 nA. In addition, Pt/C showed poor selectivity, with much stronger responses to H₂S (2170 nA) and CO (450 nA) than to NO₂ (100 nA) at 10 ppm, and much larger humidity-induced drift (from 5 to 40 nA over 33–95% RH) than CeZrO_4−δ-600 (below 5 nA). These results indicate that the higher NO₂ response and lower humidity-induced drift of CeZrO_4−δ-600 are associated with its metastable defect-engineered surface, which facilitates low-concentration NO₂ capture and interfacial charge transfer.

Fig. 2. NO₂ sensing performance and long-term operational stability of CeZrO_4−δ sensors.

a Response-current transients of the sensors upon exposure to 5 ppm NO₂ at room temperature. t_res and t_recov denote the response and recovery times. The gray region denotes the period of exposure to 5 ppm NO₂. b Continuous response and recovery curves to NO₂ in the range of 0.05-10 ppm. The inset shows the linear fit between response current and NO₂ concentration for CeZrO_4−δ and CeZrO_4−δ-600. c Response-current changes at NO₂ concentrations of 0.05, 0.1, 0.2, 0.5, 1, 5 and 10 ppm. d Long-term response retention of CeZrO_4−δ and CeZrO_4−δ-x (x = 300, 600 and 900) sensors. The dashed line marks 98% response retention. Retention was calculated relative to the initial response-current change as defined in Methods. Data are presented as mean ± standard deviation (SD) from n = 3 independently fabricated sensors per material, with individual data points shown. e Response metrics at 5 ppm NO₂ versus reported stability. Asterisks denote electrochemical current changes in nA; unmarked values are dimensionless chemiresistive response ratios. Values are compared only within the same sensing category. The corresponding data and references are provided in Table 1. Source data are provided as a Source Data file.

Table 1.

Comparison of representative NO₂ sensing materials using modality-specific response metrics

Sensing type	Sensing material	Detection range (ppm)	Response metric (5 ppm)	Reported stability (days)	Temp. (°C)	LOD (ppb)	Ref.
electrochemical sensor	CeZrO4−δ-600	0.05–20	2926	98	RT	10	This work
	In2O3 N-40min	0.1–10	350	30	RT	10	15
	TiN	0.05–50	631	180	RT	2	17
	C-SiO2-PTFE	5–100	17.5	21	RT	4300	48
chemiresistive sensor	TiOx-In2O3	1–5	35.8	14	RT	1000	49
	SnO2-SnS2-SiO2/SiO2	0.1–15	30	21	RT	25	50
	PtNR-CeO2	0.5–40	6	14	30	20	51
	SWCNT	0.01–3	41.2	90	150	0.16	52
	Au-Bix-In2-xO3	0.1–1	640	14	RT	100	53
	HIOTP-Ni	1–100	202.5	10	RT	210	54
	TiO2@NH2-MIL-125	0.1–100	150	90	RT	1	55
	MoS2/MXene	2–20	3	14	RT	500	56
	In2O3/PPy	5–30	60	21	RT	500	57

Note: Response values are modality-specific. For electrochemical sensors, the response metric is the absolute response-current change, $\mathrm{\Delta }I=\mid I_{\mathrm{gas}}-I_{\mathrm{air}}\mid$, reported in nA, where $I_{\mathrm{air}}$ and $I_{\mathrm{gas}}$ are the currents measured in air and after exposure to NO₂, respectively. For chemiresistive sensors, the response metric is the dimensionless resistance ratio reported in the original study, generally expressed as $R_{\mathrm{gas}}/R_{\mathrm{air}}$ or $R_{\mathrm{air}}/R_{\mathrm{gas}}$, according to the response direction of the sensing material. Values at 5 ppm NO₂ were taken directly from the original publications or obtained from published concentration-dependent response curves or calibration relationships when the corresponding 5 ppm value could be determined within the reported measurement range. Because electrochemical current changes and chemiresistive resistance ratios represent different physical quantities and transduction mechanisms, response magnitudes should be compared only within the same sensing category. The shared numerical scale in Fig. 2e is intended solely for descriptive visualization and does not represent a physically unified response quantity. Reported stability durations were taken from the original publications and may reflect different testing protocols and retention criteria. Limit of detection (LOD) values were taken from the original reports or, where explicitly recalculated, estimated as $\mathrm{LOD}=3\sigma /m$, where $\sigma$ is the standard deviation of the baseline noise and $m$ is the slope of the calibration curve. RT denotes room temperature.

Moreover, the sensor retained a measurable and reproducible response over the tested temperature range from − 25 to 75 °C, with the mean response-current change varying by less than 15% relative to the value at − 25 °C. This temperature insensitivity may be associated with the stable Ce³⁺/Ce⁴⁺ redox couple and the proton-conducting environment sustained by Nafion, which together help buffer thermal perturbations and maintain stable electrochemical kinetics (Supplementary Fig. S27)³⁶. To further verify that this thermal robustness is not accompanied by relaxation of the metastable phase, CeZrO_4−δ-600 powder was subjected to three thermal cycles between 25 and 75 °C, with a 30 min dwell at each temperature, followed by accelerated aging at 75 °C in air for 12 h before XRD and Rietveld analysis. The thermally treated sample retains the characteristic tetragonal reflections and a comparable tetragonal phase fraction (Supplementary Fig. S28), indicating that the metastable t-CeZrO_4−δ-rich structure does not undergo obvious relaxation to the stable cubic phase under the tested thermal-stress conditions.

The CeZrO_4−δ and CeZrO_4−δ-600 sensors maintained distinguishable response-recovery profiles over six consecutive NO₂ exposure cycles, with CeZrO_4−δ-600 showing the larger response under the tested conditions (Fig. 3a). The slow-cooled and air-cooled control sensors also showed repeated response-recovery behavior over six consecutive cycles (Supplementary Fig. S29). The CeZrO_4−δ-600 sensor shows a larger response to NO₂ than to the tested interferents under the specified concentrations (Fig. 3b). It exhibits minimal variation (< 5 nA) under varying relative humidity (RH), thanks to a compact design with a sealed water reservoir that stabilizes internal humidity (Fig. 3d). This design limits humidity-induced interference. In addition to humidity, CO₂ was evaluated as a practical exhaust-related interferent because progressive carbonate formation may deactivate oxide surfaces. The CeZrO_4−δ-600 sensor showed only minor current fluctuations upon exposure to 1–15% CO₂ alone and maintained clear responses to 1 ppm NO₂ in the presence of 1–15% CO₂ (Supplementary Fig. S30). After 24 h aging in 10% CO₂ under humidified conditions, the sensor retained 96.9% of its initial NO₂ response, while FTIR spectra showed only weak carbonate-related features that largely decreased after air purging, indicating limited irreversible carbonate accumulation under the present conditions. Additional pulse-overload tests show that brief exposure to an elevated NO₂ concentration of 500 ppm produces a large transient current response, followed by recovery toward the baseline; the CeZrO_4−δ-600 sensor remains capable of resolving subsequent 1.5 ppm NO₂ pulses (Supplementary Fig. S31). While optical sensors offer slightly faster response and lower detection limits, they are limited by higher cost, larger size, and greater power consumption (Fig. 3c)³⁷. CeZrO_4−δ-600 shows a favorable combination of the selected practical metrics at a low cost. By integrating key comprehensive factors—including cost, volume, and power—into a unified reference metric, its advantage becomes evident across the full spectrum of reported NO₂ sensors, spanning optical, mass, acoustic, and electrical platforms (Fig. 3e and Supplementary Tables S16, S17). These results indicate that metastable-polymorph engineering enhances intrinsic sensor performance while offering a favorable balance across several practical metrics relevant to next-generation NO₂ detection.

Fig. 3. Repeatability, selectivity and practical performance of CeZrO_4−δ sensors.

a Representative current-response traces of CeZrO_4−δ and CeZrO_4−δ-600 sensors during six consecutive exposure-recovery cycles with 3 ppm NO₂. Each trace was obtained from one sensor, and the six cycles represent repeated exposures of the same device without averaging. b Selectivity of CeZrO_4−δ and CeZrO_4−δ-x (x = 300, 600 and 900) sensors toward 1 ppm NO₂ and 200 ppm of each interfering gas. c Comparative overview of optical Broadband Cavity-Enhanced Absorption Spectroscopy (BBCEAS) and electrochemical NO₂ sensors in terms of cost, size, and power consumption. The BBCEAS values were obtained from ref. ³⁷. d Humidity-induced current changes of CeZrO_4−δ and CeZrO_4−δ-x (x = 300, 600 and 900) sensors in air over 33–95% RH at room temperature, without NO₂ exposure. The dashed line marks a current change of 5 nA. Data are presented as mean ± SD from n = 3 independently fabricated sensors per material, with individual data points shown. e Comparison of the limit of detection (LOD), response time and normalized comprehensive performance factor for this work and representative NO₂ sensing technologies. Large labeled symbols denote literature data, and smaller symbols denote their projections onto the coordinate planes. Source data are provided as a Source Data file.

Gas sensing mechanism

Ultraviolet-visible diffuse reflectance spectra (UV-vis DRS) show only a minor bandgap change upon quenching (ΔE_g ≈ 0.04 eV, Supplementary Fig. S32), indicating that the enhanced sensing cannot be ascribed to bandgap modulation^38,39. Instead, in situ Fourier transform infrared spectroscopy (FTIR) reveals sequential NO₂ → NO₂⁻ → NO transformations on CeZrO_4−δ-600 (Fig. 4a, b), evidencing a surface-mediated redox pathway facilitated by abundant surface-active sites^15,17. The enhanced sensing thus originates from the surfaces of the metastable t-CeZrO_4−δ stabilized by rapid quenching. Its lower symmetry relative to c-CeZrO_4−δ induces lattice distortions and creates distorted metastable tetragonal surface configurations, while the preserved Ce³⁺ population sustains abundant oxygen vacancies and adsorbed oxygen species. These features collectively promote NO₂ adsorption, accelerate charge transfer, and enhance surface-reaction kinetics. These features are particularly important under low-concentration conditions because they increase the probability that incoming NO₂ molecules are trapped and activated at the gas-facing interface before desorption, thereby promoting interfacial charge transfer and amplifying the current response. Density functional theory (DFT) calculations using idealized stoichiometric cubic and tetragonal CeZrO₄ (311) slabs show stronger NO₂ adsorption on the tetragonal (t-CeZrO₄) model than on the cubic (c-CeZrO₄) model (Supplementary Fig. S33, S34), with more negative adsorption energy (− 3.38 vs − 1.09 eV) and greater electron transfer (0.132 e vs 0.064 e, Fig. 4c–g).

Fig. 4. Spectroscopic and computational analysis of NO₂ adsorption and charge transfer on CeZrO_4−δ polymorphs.

a, b Magnified in situ Fourier transform infrared (FTIR) spectra of CeZrO_4−δ and CeZrO_4−δ-600, respectively, during exposure to 5 ppm NO₂. Spectra were collected from 0 to 70 s at 10 s intervals and vertically offset for clarity. The vertical dotted lines mark the bands at approximately 1919 and 1340 cm⁻¹, assigned to NO and NO²⁻. c, d Optimized clean and NO₂-adsorbed cubic CeZrO₄ (311) slab models. e, f Optimized clean and NO₂-adsorbed tetragonal CeZrO₄ (311) slab models. In (d, f), yellow and pink isosurfaces denote electron accumulation and depletion, respectively. g Density functional theory (DFT)-calculated adsorption energies and Bader charge transfer for NO₂ on the cubic and tetragonal CeZrO₄ surfaces. Source data are provided as a Source Data file.

Because NO₂ adsorption on t-CeZrO₄ is markedly stronger and involves larger electron transfer, the initial NO₂ capture is less likely to be the kinetically limiting step. Instead, the subsequent transformation of adsorbed nitrite- or nitrate-like species into NO, accompanied by local Ce³⁺ oxidation and recovery of the active surface, may represent a more kinetically demanding vacancy-coupled redox step. In this regard, the defect-enriched metastable t-CeZrO₄ surface may facilitate this vacancy-coupled redox step by promoting NO_x adsorption, oxygen-vacancy turnover, and interfacial charge transfer. Charge-density difference maps further show pronounced charge redistribution on the tetragonal phase, consistent with its enriched active sites. These results indicate that metastable-phase engineering is a potentially general strategy for tuning adsorption, charge transfer, and interfacial redox behavior in oxide-based electrochemical sensors.

Vehicle exhaust monitoring applications

To validate the relevance of the sensor under Euro 7 emission regulations, we conducted a near-tailpipe vehicle-exhaust monitoring test designed for compliance-oriented diagnostics. Conventional inspection and onboard systems, optimized for total NO_x evaluation, are inadequate for precise, real-time NO₂ detection under dynamic exhaust conditions—a challenge directly addressed by the CeZrO_4−δ-600 sensor. The integrated sensing platform (Fig. 5a, b and Supplementary Fig. S35, S36) featured a wireless module that enabled real-time NO₂ quantification across NO₂ concentrations of 0.1–10 ppm, achieving mean relative errors below 5% during calibration (Fig. 5c and Supplementary Fig. S37). Selectivity measurements confirmed negligible interference from major exhaust components (Fig. 5d). Under controlled laboratory conditions (27 °C, 50% RH), the system accurately quantified ~10 ppm NO₂ from collected exhaust samples (Fig. 5e). During transient engine operation (Supplementary Movie 2), the CeZrO_4−δ-600 sensor tracked dynamic NO₂ fluctuations under moderate temperature (~ 35 °C) and fluctuating humidity (~ 70% RH). Further real-vehicle testing across five consecutive ignition cycles revealed a reproducible transient NO₂ peak of approximately 15 ppm upon engine ignition, which rapidly decayed to steady-state levels (Fig. 5f). Even when the sensing distance was shortened to 5 cm from the exhaust outlet, exposing the sensor to elevated temperatures of 43–55 °C and relative humidity of 69–77%, the sensor consistently resolved dynamic NO₂ fluctuations during acceleration intervals at engine speeds of 2000, 4000, and 6000 revolutions per minute (rpm) (Fig. 5g and Supplementary Fig. S38). The wireless sensing system operated over the NO₂ concentration range relevant to Euro 7-oriented NO_x monitoring and China 6b-related exhaust diagnostics^2,40. To further improve signal quality, we applied a machine-learning (ML) denoising pipeline, which increased the signal-to-noise ratio (SNR) from 2.6 dB to 11 dB (Fig. 5h and Supplementary Fig. S39–S43). Unlike traditional smoothing techniques, which often distort key transient features, this ML approach effectively suppressed environmental noise while preserving critical signal characteristics, improving signal fidelity and interpretability. Such reliable, humidity-tolerant sensing performance under real exhaust dynamics is relevant to practical Euro 7-oriented NO_x monitoring and real-driving emissions (RDE) diagnostics, supporting metastable-polymorph CeZrO_4−δ-600 as a sensing platform for next-generation NO₂-resolved exhaust monitoring. Collectively, these findings underscore metastable engineering as a strategy to bridge materials innovation with practical emission-monitoring technologies.

Fig. 5. Wireless vehicle-exhaust monitoring and machine-learning-assisted processing of NO₂ signals.

a Schematic illustration of the wireless NO₂ detection system based on the CeZrO_4−δ-600 sensor. b Circuit hardware logic diagram of the wireless NO₂ detection system. c Calibration curve of the wireless CeZrO_4−δ-600-based system for NO₂ detection over concentrations ranging from 0.1 to 10 ppm. d Selectivity of the wireless sensing system toward 10 ppm NO₂ and 200 ppm of each interfering gas. The inset presents temperature and relative humidity recorded simultaneously. e NO₂ detection in collected vehicle exhaust using the wireless sensing system. f Real-time exhaust profiles recorded during five consecutive vehicle startups. g Time-resolved vehicle-exhaust NO₂ concentrations during acceleration intervals at engine speeds of 2000, 4000, and 6000 revolutions per minute (rpm). Shaded regions indicate accelerator operation. h Comparison of the raw, conventionally smoothed and machine-learning (ML)-processed exhaust-monitoring signals. The inset compares the signal-to-noise ratio (SNR) values of the raw and processed signals. Source data are provided as a Source Data file.

In summary, this work demonstrates a metastable-phase strategy for advanced gas sensing through the stabilization of tetragonal CeZrO_4−δ via liquid-nitrogen quenching. The non-equilibrium structure, characterized by abundant oxygen vacancies, enhanced Ce³⁺ concentration, and distorted tetragonal surface configurations, enables rapid NO₂ adsorption and electron transfer, leading to a high current response of 24,000% at 10 ppm. Compared with the equilibrium cubic counterpart, the metastable polymorph provides a higher NO₂ response while retaining stable operation during humidity, temperature and long-term tests. Beyond performance improvement, this approach establishes a versatile framework for engineering metastable systems with tailored charge dynamics. The study bridges materials thermodynamics with sensor design, offering a pathway toward sensing technologies for environmental and automotive exhaust monitoring.

Methods

Preparation of pristine CeZrO_4−δ

Pristine CeZrO_4−δ was synthesized via a modified sol-gel method. 0.01 mol of Ce(NO₃)₃·6H₂O and 0.01 mol of Zr(NO₃)₄·5H₂O were dissolved in 25 mL of deionized water under magnetic stirring. Subsequently, 0.02 mol of anhydrous citric acid was added as a complexing agent, and the mixture was stirred ultrasonically to ensure homogeneity. The solution was then heated to 120 °C until it transformed into a yellow gel. This gel was vacuum-dried at 60 °C for 10 h to remove residual water. The resulting dry gel was calcined in a muffle furnace at 900 °C for 10 h with a heating rate of 10 °C min⁻¹. The obtained pale-yellow powder was ground thoroughly and subjected to a second calcination under the same conditions to ensure phase purity.

Preparation of CeZrO_4−δ-x (x = 300, 600, 900) (liquid nitrogen-quenched samples)

To obtain samples containing the metastable t-CeZrO_4−δ phase, the as-prepared CeZrO_4−δ was placed in a ceramic crucible and heated to 300 °C, 600 °C, or 900 °C in a muffle furnace. Once the target temperature was reached, the crucible was immediately removed and quenched in liquid nitrogen until complete evaporation occurred. The resulting samples were collected and labeled as CeZrO_4−δ-300, CeZrO_4−δ-600, and CeZrO_4−δ-900, respectively. A detailed procedure is shown in Supplementary Movie 1.

Preparation of CeZrO_4−δ-x (control samples)

For comparison, two control samples were prepared. In the first case, CeZrO_4−δ was heated to 600 °C and then rapidly cooled in air by removing the crucible from the furnace. This sample was designated as CeZrO_4−δ-Q (quenched in air). In the second case, CeZrO_4−δ was heated to 600 °C and allowed to cool naturally within the furnace, yielding a sample designated as CeZrO_4−δ-S (slow-cooled).

Preparation of electrochemical gas sensor

Fuel-cell-type electrochemical gas sensors were fabricated using catalyst-coated HCP120 carbon-paper electrodes and a Nafion N115 proton-exchange membrane. A catalyst ink was prepared by ultrasonically dispersing 10 mg of sample powder, 20 μL of 5 wt% Nafion solution and 200 μL of isopropyl alcohol for 20 min. The resulting ink was divided equally between two circular HCP120 carbon-paper electrodes, with approximately 110 μL deposited on each electrode. The HCP120 carbon paper had a nominal thickness of 0.21 mm. Each electrode had a diameter of 1.0 cm and a geometric area of 0.785 cm² and contained approximately 5 mg of sample powder, corresponding to an active-material loading of approximately 6.37 mg cm⁻². Based on the nominal solid content of the Nafion dispersion, the sample powder accounted for approximately 91 wt% of the dry catalyst-layer solids. All control sensors were fabricated using the same procedure, electrode dimensions and active-material loading.

After deposition, the catalyst-coated carbon-paper electrodes were dried at 60 °C for 10 min. A Nafion N115 membrane with a nominal dry thickness of 127 μm was placed between the two catalyst-coated carbon-paper electrodes. The resulting membrane-electrode assembly was hot-pressed at 90 °C and 1 MPa for 90 s. A polytetrafluoroethylene (PTFE) film was incorporated as a protective layer. The assembly was combined with two circular copper current collectors, each 1.0 cm in diameter, and sealed using hot-melt adhesive. The copper components served only as current collectors. A small chamber containing cotton saturated with deionized water was attached to the anode side of the sensor. During operation, the cathode was exposed to the target atmosphere, whereas the anode remained sealed in a humidified-air environment (Supplementary Fig. S44). The Nafion membrane transported protons while blocking electron conduction, forcing electrons generated by the interfacial redox reactions to pass through the external circuit and produce the measured current signal.

Gas-sensing measurements and data analysis

Gas-sensing measurements were performed using a multichannel potentiostat (CHI1030C, Shanghai Chenhua Instrument Co., Ltd.) in i-t mode at an applied potential of 0 V. Unless otherwise specified, measurements were conducted at 22 ± 2 °C and 45 ± 3% relative humidity (RH). Target atmospheres were prepared in a 250 mL static gas-mixing chamber using a certified 200 ppm standard gas mixture balanced with N₂. A defined volume ($V_{inj}$) of the standard gas was injected with a gas-tight syringe, while an equal volume of the chamber gas was simultaneously vented to maintain constant total volume. The mixture was allowed to equilibrate for 1 min before measurement. The target analyte concentration was calculated as

$$C_{target}=C_{std}\times \frac{V_{inj}}{V_{ch}}$$ 1

where $C_{target}$ (ppm) is the target concentration in the chamber, $C_{std}$ (ppm) is the certified concentration of the standard gas mixture (200 ppm), $V_{inj}$ (mL) is the injected volume of the standard mixture, and $V_{ch}$ (mL) is the chamber volume (250 mL). Air was used as the background gas for all sensing measurements.

The response-current change was defined as

$$\Delta I=\mid I_{\text{gas}}-I_{\text{air}}\mid$$ 2

where $I_{\text{air}}$ and $I_{\text{gas}}$ are the currents measured in air and after exposure to the target gas, respectively. The percentage sensing response was calculated as

$$R=\frac{\mid I_{\text{gas}}-I_{\text{air}}\mid }{I_{\text{air}}}\times 100\%$$ 3

Long-term stability was evaluated by repeatedly exposing the same three independently fabricated sensors of each material to 5 ppm NO₂. For sensor i, the response retention at elapsed time t was calculated as

$${\eta }_{i,t}=\frac{{\Delta I}_{i,t}}{{\Delta I}_{i,0}}\times 100\%$$ 4

where ${\Delta I}_{i,t}$ and ${\Delta I}_{i,0}$ are the response-current changes of sensor i at elapsed time t and at the beginning of the stability test, respectively. The retention values were calculated separately for the three sensors and were subsequently used to determine the mean and standard deviation at each time point. The same sensors were followed throughout the 98-day stability test. The response and recovery times, t₉₀, were defined as the times required to reach 90% of the total current change after NO₂ introduction and removal, respectively. The sensitivity was obtained from the slope of the linear fit between ΔI and NO₂ concentration.

Material characterization

Time-of-flight (TOF) neutron powder diffraction (NPD) data were collected on 2 g of CeZrO_4−δ-600 using the General-Purpose Powder Diffractometer (GPPD) at the China Scattering Neutron Source (CSNS, Dongguan, China). The phase information and lattice parameters of prepared sensor materials were characterized using X-ray diffraction (XRD, Rigaku MiniFlex 600 powder X-ray diffractometer with Cu Kα radiation at λ = 1.5418 Å, scanning speed: 0.1 ° min⁻¹, 2θ range of 10–90°). Scanning electron microscopy (SEM; JSM-7900F, Japan) was used to observe the surface morphology. The high-resolution transmission electron microscopy (HRTEM) images were acquired on a JEM-F200 (JEOL Ltd., Japan) instrument. X-ray photoelectron spectroscopy (XPS) was performed using a K-Alpha + (USA) instrument. The binding energies in the XPS spectra of CeZrO_4−δ and CeZrO_4−δ-x (x = 300, 600, 900) were calibrated using the C 1 s peak at 284.8 eV. The Ce L₃-edge X-ray absorption fine structure (XAFS) spectra were collected at beamline BL13SSW of the Shanghai Synchrotron Radiation Facility (SSRF, Shanghai, China), whereas the Zr K-edge XAFS spectra were recorded in transmission mode using an easyXAFS 300 spectrometer (easyXAFS, USA). Brunauer-Emmett-Teller (BET) results of CeZrO_4−δ and CeZrO_4−δ-x (x = 300, 600, 900) were obtained by an automatic specific surface area and microporous pore analyzer (3FLEX, USA). UV-vis diffuse reflectance spectra (UV-vis-DRS) were measured using a UV-550 spectrophotometer. In situ infrared spectroscopy (in situ IR) was carried out using a VERTEX 70 spectrometer equipped with an MCT detector; all spectra were recorded by co-adding 32 scans at a resolution of 4 cm⁻¹ over 4000-400 cm⁻¹.

DFT calculations

DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP, version 5.4.4) within the projector-augmented-wave (PAW) framework^41,42. Exchange-correlation interactions were described using the Perdew-Burke-Ernzerhof (PBE) functional within the generalized gradient approximation⁴³. Long-range dispersion interactions were accounted for using Grimme’s DFT-D3 method with Becke-Johnson damping⁴⁴. To compare the intrinsic phase-dependent adsorption behavior, stoichiometric slab models exposing the (311) surfaces of cubic (Fm3̅m) and tetragonal (P4₂/nmc) CeZrO₄ were used as idealized representations of the corresponding experimental CeZrO_4−δ phases.

For the slab calculations, the Brillouin zone was sampled using a 4 × 4 × 1 Monkhorst-Pack k-point mesh. A vacuum region of 15 Å was introduced along the surface-normal direction to minimize interactions between periodically repeated slabs. The plane-wave kinetic-energy cutoff was set to 450 eV. The atoms in the upper two layers of each slab were allowed to relax, whereas all remaining atoms were fixed at their bulk positions. Structural relaxation was continued until the residual force on each relaxed atom was below 0.01 eV Å⁻¹, and the electronic self-consistent-field convergence threshold was set to 1 × 10⁻⁴ eV. The isolated NO₂ molecule was treated using Γ-point-only (1 × 1 × 1) sampling.

The NO₂ adsorption energy, Eads, was calculated as

$$E_{\mathrm{ads}}=E_{{\mathrm{slab}+\mathrm{NO}}_2}-E_{\mathrm{slab}}-E_{{\mathrm{NO}}_2}$$ 5

where $E_{{\mathrm{slab}+\mathrm{NO}}_2}$, $E_{\mathrm{slab}}$ and $E_{{\mathrm{NO}}_2}$ are the total energies of the NO₂-adsorbed slab, the corresponding clean slab and the isolated NO₂ molecule, respectively. Negative E_ads values indicate exothermic adsorption.

Electron transfer between the adsorbed NO₂ molecule and the CeZrO₄ slab was quantified using Bader charge analysis⁴⁵. The charge-density difference, Δρ, was calculated as

$$\Delta \rho ={\rho }_{{\mathrm{slab}+\mathrm{NO}}_2}-{\rho }_{\mathrm{slab}}-{\rho }_{{\mathrm{NO}}_2}$$ 6

where ${\rho }_{{\mathrm{slab}+\mathrm{NO}}_2}$, ${\rho }_{\mathrm{slab}}$ and ${\rho }_{{\mathrm{NO}}_2}$ are the charge densities of the NO₂-adsorbed slab, the clean slab and the isolated NO₂ molecule, respectively. Positive and negative values of Δρ represent electron accumulation and depletion, respectively. The charge-density-difference isosurfaces were plotted at ± 0.013 |e| Å⁻³ for both the cubic and tetragonal CeZrO₄ models. The lattice vectors and atomic coordinates of the optimized isolated NO₂ molecule and the clean and NO₂-adsorbed cubic and tetragonal CeZrO₄ (311) slab models are provided in Supplementary Data 3.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

X.M. conceived the project, designed the experiments, performed data analysis, and wrote the manuscript. J.L. and F.L. contributed to machine-learning-assisted signal processing. H.Z., J.H., Z.Z., C.H., and C.Y. contributed to materials synthesis, characterization, and sensor measurements. J.Z. contributed to the synchrotron measurements. J.P.A. and M.Y. supervised the project, secured funding, and revised the manuscript. All authors discussed the results and commented on the manuscript.

Peer review

Peer review information

Nature Communications thanks Gu Xiuquan, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the Young Scientists Fund (Class A) of the National Natural Science Foundation of China (Grant No. 22525102), the National Natural Science Foundation of China (Grant No. 62471085), the Fundamental Research Funds for the Central Universities (Grant No. DUT25YG103), the Program of Introducing Talents of Discipline to Universities (Grant No. B13012), and the Engineering and Physical Sciences Research Council (EPSRC, UK).

Data availability

The data supporting the findings of this study are available within the Article, Supplementary Information, Source Data and Supplementary Data 1–3. Supplementary Data 1 contains the crystallographic information files corresponding to the refined structures, Supplementary Data 2 contains the raw and fitted XPS data and associated fitting parameters, and Supplementary Data 3 contains the optimized atomic structures of the isolated NO₂ molecule and the clean and NO₂-adsorbed cubic and tetragonal CeZrO₄ (311) slab models used in the DFT calculations. The crystallographic data have also been deposited at the Cambridge Crystallographic Data Centre under deposition numbers CCDC 2575619–2575632 and can be obtained free of charge via https://www.ccdc.cam.ac.uk/structures/. Source data are provided with this paper and are also available in the Figshare repository⁴⁶ (https://doi.org/10.6084/m9.figshare.33110843). Source data are provided in this paper.

Code availability

The custom code used for ML-assisted processing of the exhaust-monitoring signals is publicly available in Zenodo under https://doi.org/10.5281/zenodo.21506873. The archived release is available without access restrictions under the MIT licence⁴⁷.

Competing interests

F.L. is an employee of Google LLC and contributed to the machine-learning-assisted signal processing described in this work. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77644-4.