Surface Terminations of LaAlO₃ Perovskite Nanoparticles as Viewed by Solid-State Nuclear Magnetic Resonance

Abstract

Nanocrystal surfaces generally undergo reconstructions that differentiate them from the bulk structures, often in nontrivial ways. Understanding these terminations is critical across diverse fields, from heterogeneous catalysis to the formation of topological states and the synthesis of semiconductor nanomaterials. Determining surface structures is currently an interdisciplinary task, most often involving high-resolution electron microscopy and surface electron diffraction. These methods, however, do not provide a global view of the ensemble of structures present in a sample. Here, we show how surface-sensitive solid-state nuclear magnetic resonance (SSNMR) spectroscopy methods can bridge this gap. In this context, we investigated the surface structure of lanthanum aluminate (LaAlO₃) perovskite nanoparticles. Four distinct surface terminations have previously been observed for this material, but their relative abundances were unknown. Using an array of double- and triple-resonance SSNMR methods probing the relative proximities of surface ¹H, ²⁷Al, ¹⁷O, and ¹³⁹La nuclei, we conclude the surface to be majority terminated (80%) by AlO _x with substantial (20%) LaO _x terminated regions.

1. Introduction

Despite the high degree to which the bulk structures of crystalline materials are now understood, following over a century of crystallography, the surfaces of crystals remain largely a hinterland. It is often falsely assumed that crystalline materials undergo bulk truncations that mirror the periodic structure, but the reality is far more unsettling. Surfaces generally undergo complex reconstructions that may introduce new functionalities that are absent from the bulk and are, importantly, not unique. , For example, 30 distinct surface terminations have been observed experimentally for SrTiO₃, − easily the material with the most extensively studied surface chemistry. This, of course, excludes nonperiodic termination structures that may be present in even larger quantities.

Being able to elucidate and quantify surface terminations of crystalline materials is an important frontier. Surface functionalities have direct implications for heterogeneous catalysis, for instance, where reconstructions can introduce Lewis or Brønsted acid sites that are absent from the bulk or necessary for metal-ion immobilization. − Surface structures are also implicated in the formation of trap states in semiconductor nanomaterials which hinder their luminosity and photovoltaic properties. − Surfaces and interfaces are also relevant for the formation of superconducting states, for example, at the interface of SrTiO₃ and LaAlO₃. −

Thus, far, the subject of surface termination has been largely approached from two directions. The first approach is from a purely computational posture. Computer programs such as MAGUS, and USPEX, attempt to predict the most likely surface termination for a given facet and supercell through algorithm-based structure searches. While these tools are critical for postulating possible structures in underexplored materials and provide better alternatives to bulk truncations, experimental evaluations of surface structures are needed. Electron microscopy and surface electron diffraction are the primary methods used to determine surface structures. , High-resolution electron microscopy provides direct visualization of surface morphology and defect structures, while electron diffraction offers complementary information on the surface’s long-range order and crystallographic orientation. These tools have been used to great success for determining the structures of some 2D periodic surfaces.

While this toolkit has enabled exquisite detail in experimentally determined facet structures in certain materials, chiefly perovskite oxides, there are obvious limitations. Microscopy methods characterize small regions of the surface, often with a bias toward focusing on the more well-defined and crystalline facets. Consequently, characterizing the entire surface is time prohibitive, and even then, remains blind to the significant fractions of the surfaces that could be disordered or dynamic. Therefore, alternative methods capable of viewing the entire surface as an ensemble are essential for establishing accurate structure–property relationships.

Solid-state nuclear magnetic resonance (SSNMR) spectroscopy has been used for decades to characterize surfaces. − The method is particularly applicable to the study of disordered surfaces because, as an ensemble-average method, it is equally sensitive to both periodic and amorphous solids. Surface sites are nevertheless dilute, which leads to low sensitivity. Common approaches used for the detection of surface sites include polarization transfer and indirect detection methods, for instance through abundant ¹H spins. The advent of dynamic nuclear polarization (DNP) , surface-enhanced NMR spectroscopy (SENS) , has revitalized the use of SSNMR for surface site detection, often leading to atomistic-level structural resolutions. −

Herein, we applied sensitivity-enhanced SSNMR methods to probe surface structures across the entire surface of lanthanum aluminate (LaAlO₃) nanoparticles. LaAlO₃ surfaces are relevant in heterogeneous catalysis and for the formation of topological superconducting states. − Several different terminations have been observed experimentally for this material , that can be categorized into four general families (Figure ). Namely, the particle can form an AlO _x monolayer, an LaO _x monolayer, an AlO _x bilayer, or a mixed-metal termination. We sought to resolve which of these structures are present in the highest quantities, and whether their relative abundances could be estimated. Importantly, the tools developed as a part of this work enable the evaluation of how surface termination is impacted by external forces or synthetic approaches, which is challenging to gauge otherwise.

1.

Surface terminations of LaAlO₃ obtained from prior electron microscopy and surface electron diffraction investigations. , The structures are representative of an (A) AlO _x monolayer, (B) LaO _x monolayer, (C) AlO _x bilayer, and (D) mixed LaAlO _x layer. Colored spheres represent H (white), O (red), Al (blue), and La (green) atoms.

2. Materials and Methods

2.1. Synthesis of LaAlO₃

LaAlO₃ nanoparticles were prepared via a two-line hydrosauna synthesis modified from a literature procedure for LaScO₃ nanoparticles. , An equimolar solution of La³⁺ and Al³⁺ was prepared by dissolving La­(NO₃)₃·6H₂O (2.5 mmol, 1.082 g; Alfa Aesar, 99.9%) and Al­(NO₃)₃·9H₂O (2.5 mmol, 0.938 g; Fisher Chemical, 98.0 to 102.0%) in deionized water (25 mL). Concentrated NH₄OH solution (50 mL; Fisher Chemical, 28.0 to 30.0% w/w) was preheated and stirred at 60–70 °C for 5–10 min to partially vent ammonia. The metal-ion solution was then added dropwise to the hot base solution, precipitating a mixed-cation hydrogel. The hydrogel was collected by centrifugation (7000 rpm, 7 min), redispersed, and washed with deionized water three times. The wet hydrogel was gently crumbled and loosely spread in an alumina boat, leaving gaps to enable gas flow. The material was then calcined in a tube furnace under a controlled, flowing, humid environment generated upstream of the furnace by mixing dry Ar gas and saturated Ar gas flowing through a water bubbler, setting a total inlet water vapor partial pressure of 1 Torr. Under this two-line hydrosauna reactor, the hydrogel was held at 700 °C for 48 h, followed by 400 °C for 24 h.

2.1.1. ¹⁷O-Surface Enrichment of LaAlO₃

The LaAlO₃ nanoparticles were surface-enriched with ¹⁷O by following analogous procedures reported in the literature. The LaAlO₃ nanoparticles were heated to 300 °C for 8 h under vacuum, wetted with 39.3% ¹⁷O-labeled H₂O (Cortecnet) and left overnight. The nanoparticles were then dried under vacuum at room temperature for 8 h.

2.2. Powder X-ray Diffraction (PXRD)

Powder X-ray diffraction measurements were acquired at room temperature using a Rigaku Ultima diffractometer featuring a Cu Kα radiation source. The PXRD patterns were acquired by scanning 2θ from 10° to 70° with a step size of 0.05°.

2.3. Electron Microscopy and Energy-Dispersive X-ray Spectroscopy

Transmission electron microscopy (TEM) was performed on a JEOL JEM-2100 microscope operating at an accelerating voltage of 200 kV to investigate the morphology and elemental distribution of the nanoparticles. Energy-dispersive X-ray spectroscopy (EDS) was performed in scanning transmission electron microscopy (STEM) mode for compositional analysis. For TEM preparation, 0.2 mg of LaAlO₃ was dispersed in 200 μL of ethanol and sonicated for 30 min. The resulting suspension was drop-cast onto a 200-mesh carbon-coated copper grid and dried under ambient conditions. The Cu and C signals observed in the EDS spectra are therefore attributed to the carbon-coated copper support grid rather than the LaAlO₃ sample.

2.4. Solid-State NMR (SSNMR)

DNP-enhanced SSNMR experiments were acquired using a Bruker Avance III 400 MHz/263 GHz MAS-DNP NMR spectrometer equipped with a Bruker 3.2 mm low-temperature triple-resonance MAS probe. For these measurements, the sample was impregnated with a 16 mM solution of the TEKPol polarizing agent in perdeuterated 1,1,2,2-tetrachloroethane. −

Conventional SSNMR experiments were instead performed using a Bruker Avance III 600 MHz NMR spectrometer equipped with a Varian 3.2 mm triple-resonance MAS probe, or a JEOL 0.75 mm double-resonance fast-MAS probe with the exception of the ¹⁷O NMR spectrum, which was acquired using an Agilent DD2 400 MHz NMR spectrometer. Spectra and dipolar dephasing data were processed and simulated using SIMPSON, , INTERFACES, and ssNake V1.5. Chemical shifts were referenced using the universal shielding scale to TMS (δ_iso(¹H) = 0 ppm) using the CH₂ peak of adamantane (δ_iso (¹H) = 1.82 ppm) as a secondary reference.

2.4.1. ¹H SSNMR

2.4.1.1. ¹H­{²⁷Al} and ¹H­{¹³⁹La} RESPDOR

Rotational-echo saturation-pulse double-resonance (RESPDOR) , experiments were performed at 14.1 T using a 0.75 mm rotor and a MAS frequency of 71.419 kHz corresponding to a rotor period (t _r) of 14 μs. The ¹H π/2 and π pulse durations were 1.75 μs, and 3.5 μs, respectively. SR4₁ ² heteronuclear dipolar recoupling was applied to reintroduce the ¹H–²⁷Al or ¹³⁹La dipolar interactions. The recoupling time for the RESPDOR experiments was incremented from 2t _r to 178t _r in 2t _r increments. ²⁷Al and ¹³⁹La saturation pules were applied for 21 μs, corresponding to 1.5t _r, and the RF amplitudes were optimized to provide maximum dephasing. Spectra were acquired in 256 scans with a 1 s recycle delay. The ¹H­{²⁷Al} and ¹H­{¹³⁹La} RESPDOR data were simulated using the INTERFACES program using structural models of the four possible surface terminations (Figure ) that considered all the atomic sites within 1.5 nm of the surface H atom(s). Owing to spurious recoupling of ¹H chemical shift anisotropy and unavoidable rotor instabilities, , decay is observed at longer recoupling times. Uncertainties (2σ­(t)) in the RESPDOR data were calculated by the following expression:

$$2\sigma (\mathrm{t})=2\frac{\mathrm{\Delta }S(t)·N}{S_0{(t)}^2}$$

where ΔS(t) and S ₀(t) are the dipolar difference and reference ¹H signal intensities at recoupling time t and N is the noise.

2.4.2. ¹⁷O SSNMR

2.4.2.1. ¹⁷O­{¹H} QCPMG

A room temperature ¹⁷O SSNMR spectrum of the ¹⁷O-surface-enriched LaAlO₃ nanoparticles was acquired at 9.4 T using the quadrupolar Carr–Purcell–Meiboom–Gill (QCPMG) pulse sequence. The spectrum was acquired at an MAS frequency of 11.111 kHz and signal averaged over 131,072 scans using 10 and 20 μs ¹⁷O central transition-selective π/2 and π pulses, 25 kHz ¹H decoupling, a spikelet spacing of 1.852 kHz (74 echoes), and a 1 s recycle delay.

2.4.2.2. ¹H→¹⁷O PRESTO-QCPMG

A DNP-enhanced ¹⁷O SSNMR spectrum of the ¹⁷O-surface-enriched LaAlO₃ nanoparticles was acquired using the phase-shifted recoupling effects a smooth transfer of order (PRESTO-II) , QCPMG pulse sequence at an MAS frequency of 11.111 kHz. The spectrum was acquired in 256 scans using 5 and 10 μs ¹⁷O central transition-selective π/2 and π pulses, 100 kHz ¹H decoupling, a spikelet spacing of 1.111 kHz (54 echoes), and a 3 s recycle delay. A total of two rotor periods of R18₁ ⁷ recoupling were applied with a ¹H RF amplitude of 100 kHz.

2.4.2.3. ¹H→¹⁷O­{²⁷Al} PRESTO-TRAPDOR-QCPMG

A DNP-enhanced PRESTO-II-transfer of population in a double-resonance (TRAPDOR) , -QCPMG experiment was performed on the ¹⁷O-surface-enriched LaAlO₃ nanoparticles using the above parameters. ²⁷Al TRAPDOR dephasing was incremented from 2t _r to 112t _r in 80 μs increments with an ²⁷Al RF amplitude optimized to provide the maximum dephasing. Sixteen scans were acquired for each spectrum. The ¹⁷O­{²⁷Al} TRAPDOR dephasing curves were simulated using SIMPSON. , The simulation used ¹⁷O and ²⁷Al quadrupolar coupling constants (C _Q) of 4.6 and 3.8 MHz, respectively, corresponding to their average values yielded from experiment (vide infra, Table S1). Powder averaging was accomplished using a 2000-angle REPULSION set and 4 gamma angles. Effective ¹⁷O–²⁷Al dipolar coupling constants of 468, 653, 970, and 1131 Hz were used for the LaO _x monolayer, mixed LaAlO _x layer, AlO _x monolayer, and AlO _x bilayers. These values correspond to the root-sum-squares dipolar coupling constants calculated from the surface hydroxyl O sites and all the Al sites within a 1.5 nm radius for the given model. Vertical scaling factors were applied to the simulated dephasing curves to obtain a reasonable match to the experimentally measured dephasing plateau. Uncertainties in the TRAPDOR dephasing curves were calculated following an analogous procedure as outlined in §2.4.1.

2.4.2.4. ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-HETCOR-TRAPDOR-QCPMG

DNP-enhanced ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-heteronuclear correlation (HETCOR)-TRAPDOR-QCPMG spectra were acquired, with and without 12t _r of ²⁷Al dephasing using the parameters listed above. The ¹H chemical shift evolution period was incremented in 32 56.64 μs increments, for an indirect spectral width of 44.0907 ppm. Frequency-switched Lee–Goldberg (FSLG) homonuclear decoupling with a ¹H RF amplitude of 100 kHz was applied to improve the resolution in the ¹H dimension.

2.4.3. ²⁷Al SSNMR

2.4.3.1. ²⁷Al Hahn Echo

An ²⁷Al SSNMR spectrum was acquired at 14.1 T using the Hahn echo pulse sequence at room temperature. A 3.2 mm triple-resonance probe was used with a MAS frequency of 20 kHz. The spectrum was acquired in 1024 scans using 5 and 10 μs central-transition selective π/2 and π pulses and a recycle delay of 5 s.

2.4.3.2. ¹H→²⁷Al PRESTO-III

A DNP-enhanced ¹H →²⁷Al SSNMR spectrum was acquired using the PRESTO-III pulse sequence on the ¹⁷O-surface-enriched LaAlO₃ nanoparticles using a 3.2 mm triple-resonance probe and a MAS frequency of 11.111 kHz. The spectrum was acquired using π/2 and π CT-selective pulses with lengths of 5 and 10 μs, respectively, a ¹H decoupling radiofrequency field of 100 kHz, a recycle delay of 6 s, and averaged using 4096 scans. 12t _r of R18₁ ⁷ recoupling were applied.

2.4.3.3. ²⁷Al DQ-SQ

A ²⁷Al double quantum–single quantum (DQ-SQ) spectrum was acquired at 14.1 T using a 3.2 mm triple-resonance probe and 100 kHz ¹H decoupling. The spectrum was recorded using 32t _r of BR2₁ ² recoupling with a 10 μs ²⁷Al central transition selective π/2 pulse, a z-filter of 250 μs, and a 20 kHz MAS frequency. The recycle delay was set to 0.25 s, and 4800 scans were acquired for each of the 68 30 μs t ₁ increments.

2.4.4. ¹³⁹La SSNMR

2.4.4.1. ¹³⁹La QCPMG

A ¹³⁹La SSNMR spectrum was acquired at 14.1 T using the QCPMG pulse sequence at room temperature, a 3.2 mm triple-resonance probe, and a MAS frequency of 20 kHz. ¹³⁹La CT-selective π/2 and π pulse lengths were 2.5 and 5 μs, respectively. A spikelet spacing of 1 kHz (49 echoes) and a recycle delay of 2 s were used. A total of 1024 scans were acquired.

2.4.4.2. ¹H→¹³⁹La D-RINEPT-SR4₁ ²(tt)-QCPMG

A DNP-enhanced ¹H→¹³⁹La D-RINEPT- SR4₁ ²(tt)-QCPMG , spectrum was acquired using the variable-offset cumulative spectra (VOCS) approach. This approach was needed owing to the much larger quadrupolar interactions of the surface sites when compared to the near cubic environments of the bulk sites. The MAS rate was set to 12.5 kHz and ¹³⁹La π/2 and π pulse lengths were set to 7.5 and 15 μs, respectively, with a spikelet spacing of 3125 Hz, a recycle delay of 13 s, and 80 kHz ¹H decoupling. The adiabatic SR4₁ ²(tt) pulses used a ¹H RF amplitude of 125 kHz, and a shaped pulse profile with a sweep width of 4 MHz, a peak amplitude parameter of 20, and a frequency modulation parameter of 10. The ¹H square pulses used a RF power of 100 kHz. VOCS spectra were collected by stepping the transmitter by 31.25 kHz from 218.75 kHz to −218.75 kHz for a total of 21 subspectra. Each subspectrum consisted of 128 scans.

2.5. Density Functional Theory Calculations

Plane-wave density functional theory (DFT) calculations were performed using CASTEP in Materials Studio 2018. Geometry optimizations and NMR tensor calculations employed the PBE functional with a planewave energy cutoff of 630 eV, a k-point spacing of 0.07, on-the-fly generated ultrasoft pseudopotentials, and scalar relativistic effects treated using the zeroth-order regular approximation (ZORA). Convergence was reached using a maximum change in energy of 5.0 × 10^–6 eV atom^–1, in force of 0.01 eV Å^–1, in stress of 0.02 GPa, and in displacement of 5.0 × 10^–4 Å. Nuclear electric quadrupole moments of – 25.58, 146.6, and 206 mb were used for ¹⁷O, ²⁷Al, and ¹³⁹La, respectively. Calculated ¹⁷O, ²⁷Al, and ¹³⁹La magnetic shielding values were converted to chemical shifts using the calculated magnetic shielding values from the R3̅c crystal structure of LaAlO₃. This corresponded to absolute shielding values of 76.18, 545.14, and 4543.39 ppm for ¹⁷O, ²⁷Al, and ¹³⁹La, respectively.

3. Research and Discussion

In the following sections, we discuss the morphology and bulk of the LaAlO₃ nanoparticles (§ ) and finally the sensitivity-enhanced NMR investigations of the nanoparticles (§ ). Surface-selective internuclear dipolar coupling measurements are compared directly to predictions from the surface terminations observed in references Kienzle et al., and Lanier et al., and to combinations of these structures to gain a complete picture of the mean surface termination of the particles.

3.1. LaAlO₃ Nanoparticles

The morphology and bulk structure of the LaAlO₃ particles were characterized by PXRD, TEM, and EDS (Figure ). The PXRD pattern revealed a crystalline bulk corresponding to LaAlO₃ (Figure A), however, additional peaks observed between 27° and 30° indicate a La₄Al₂O₉ component (second phase) of ∼10 wt %, as estimated by the relative intensities of the most intense peak observed between the two phases. The TEM images revealed the LaAlO₃ sample consists of nanoparticles with a broad distribution of shapes and sizes (Figures and S1), and the EDS elemental mapping images showed that the LaAlO₃ phases are dopant-free (Figures and S2). As also depicted in Figure at a higher magnification, the particles are well-defined and exhibit significant relatively sharp-edged facets. We also observed additional phases in the EDS mapping images that were not detected by PXRD (Figures S2 and S3), suggesting that they are rather minor in abundance.

2.

(A) PXRD patterns, (B, C) TEM images, and (D–F) EDS mapping images of LaAlO₃ nanoparticles. (A) The experimental PXRD pattern of LaAlO₃ is in black, and simulated PXRD patterns of LaAlO₃ and La₄Al₂O₉ are in red. The EDS mapping images parts D, E, and F correspond to elemental maps of O, Al, and La, respectively. The white background in part E is shown to increase the contrast for the red color representing Al.

To address whether these additional phases would hinder the upcoming SSNMR analysis, the crystal structures of the dominate phases LaAlO₃ and La₄Al₂O₉ were obtained from the Materials Project structure database (MP-2920 and MP-781707, respectively) and used as starting points for calculating their NMR tensors (Table S1). Simulated ²⁷Al and ¹³⁹La SSNMR spectra using those calculated NMR tensor values and the weight-percentage from the PXRD data, revealed that the La₄Al₂O₉ phase would not complicate the NMR analysis (Figures S4 and S5). In this context, ²⁷Al and ¹³⁹La SSNMR spectra were acquired (Figure ). The ²⁷Al SSNMR spectrum contained three signals, corresponding to four-, five-, and six-coordinate Al (Al^IV, Al^V, Al^VI). To aid in the assignment, a ²⁷Al DQ-SQ SSNMR spectrum was acquired (Figure B). Correlations were observed between the Al^VI and Al^IV resonances suggesting that those sites are spatially proximate. Prior literature by Kentgens et al., attributed this Al^IV site to surface sites. The relatively high abundance is consistent with the high surface-to-bulk ratios of these small nanoparticles. The Al^V site may originate from the La₄Al₂O₉ phase, however, it is important to note that some LaAlO₃ termination structures include Al^V. Regardless, the signal represents a minor species. The ¹³⁹La SSNMR spectrum (Figure C) revealed a powder pattern that was best reproduced using a Czjzek model with an average C _Q of 8 MHz, which is consistent with NMR tensor calculations from the LaAlO₃ crystal structure (Table S1). The use of the Czjzek model is justified as the amorphous outer layers lead to a distribution of ¹³⁹La environments. The ¹³⁹La SSNMR signals from the La₄Al₂O₉ phase were not observed; these are low in weight-percentage and are expected to possess relatively large C _Q values (Figure S5), which diminishes its spectral intensity and severely broadens its powder pattern. Collectively, these data provide strong evidence that the sample contains mostly LaAlO₃ nanoparticles that feature a crystalline core and a potentially disordered surface.

3.

SSNMR spectra of LaAlO₃: (A) ²⁷Al Hahn echo, (B) ²⁷Al DQ-SQ, and (C) ¹³⁹La QCPMG. Spectral fitting of the ¹³⁹La SSNMR spectrum is in red. Roman numerals in (A) are used to refer to the aluminum coordination numbers.

3.2. Surface Sites on LaAlO₃ Nanoparticles

DNP SENS data were acquired on the LaAlO₃ nanoparticles after ¹⁷O-surface enrichment (§2.1) to probe the local Al, La, and O atomic environments at the surface. To detect surface ²⁷Al sites, we applied the ¹H→²⁷Al PRESTO-III method which has been shown to excite surface ²⁷Al sites with a relatively high uniformity. , The spectrum (Figure A) features resonances from Al^IV, Al^V, and Al^VI surface sites, suggesting that all three are present at the surface.

4.

DNP-enhanced SENS spectra of LaAlO₃ acquired using (A) ¹H→²⁷Al PRESTO and (B) ¹H→¹³⁹La D-RINEPT-SR4₁ ²(tt)-QCPMG. Roman numerals in (A) are used to refer to the aluminum coordination numbers.

To detect surface ¹³⁹La sites, we instead applied the ¹H→¹³⁹La D-RINEPT-SR4₁ ²(tt)-QCPMG experiment , which we recently applied together with a VOCS approach to study LaScO₃. The spectrum we measured for the LaAlO₃ nanoparticles is shown in Figure B and features a broad lineshape that is consistent with a distribution of La environments.

The ¹H→¹⁷O PRESTO-QCPMG spectrum is shown in Figure A and contains a broad resonance from all the surface hydroxyls found in the material. Note that owing to dipolar truncation, PRESTO-II is uniquely sensitive to hydroxyls and does not excite oxide sites from the bulk. A room temperature direct excitation ¹⁷O QCPMG spectrum was also acquired (Figure S6). This spectrum featured an analogous signal from surface hydroxyls with the addition of a shoulder from the bulk oxide sites at ∼ 170 ppm, as previously observed, and some increased intensity to higher frequency likely originating from surface or subsurface oxide sites; however, no La-oxide signals were detected. , Because the signals arising from the surface hydroxyls may contain distinct signals from Al–OH and La–OH species from different surface reconstructions, we acquired a 2D ¹H→¹⁷O­{¹H} PRESTO-HETCOR-QCPMG spectrum (Figure B) to improve the resolution. This spectrum features an asymmetric correlation map for the ¹H and ¹⁷O spin pairs at the surface suggesting that multiple sites may be present; however, they were not resolved.

5.

¹⁷O-DNP SENS spectra of LaAlO₃ nanoparticles: (A) ¹H→¹⁷O­{¹H} PRESTO-QCPMG and ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-HETCOR-TRAPDOR-QCPMG (B) without and (C) with a ²⁷Al dephasing pulse. (D) The difference spectrum between the ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-HETCOR-TRAPDOR-QCPMG without and with ²⁷Al dephasing, referred to as the Al–OH spectrum. (E) The difference spectrum between the ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-HETCOR-TRAPDOR-QCPMG without ²⁷Al dephasing and the Al–OH spectrum, referred to as the La–OH spectrum. Spectral simulations are shown in red, and dotted red lines are the deconvoluted simulations. Asterisk indicates spinning side bands.

To disentangle these sites, we performed a modified HETCOR experiment wherein dipolar dephasing through ²⁷Al was introduced via the TRAPDOR method. The resulting ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-HETCOR-TRAPDOR-QCPMG spectrum is depicted in Figure C and is expected to have a lower contribution from Al–OH hydroxyls, given their strong ¹⁷O–²⁷Al dipolar interactions. Therefore, by subtracting the ¹H→¹⁷O­{¹H, ²⁷Al} PRESTO-HETCOR-TRAPDOR-QCPMG spectrum (S, Figure C) from the reference ¹H→¹⁷O­{¹H} PRESTO-HETCOR-QCPMG spectrum (S ₀, Figure B), a ¹H→¹⁷O­{¹H} HETCOR spectrum was produced featuring only signal intensity from ²⁷Al–¹⁷O-¹H sites (Figure D). This difference spectrum produced a notably different, and narrower, ¹⁷O NMR lineshape, confirming the multisite nature of the spectrum and that La–OH species must also be present at the surface.

We were unable to perform ¹⁷O­{¹³⁹La} TRAPDOR to obtain a La–OH spectrum directly because of the two isotopes’ similar resonance frequencies. Difference spectroscopy could nevertheless be used to produce a La–OH ¹H→¹⁷O­{¹H} HETCOR spectrum by subtracting the aluminol spectrum from the reference spectrum. To achieve this, a vertical scaling was applied to visually cancel the Al–OH correlation from the HETCOR spectrum. While this is highly qualitative, the resulting spectrum, which is expected to feature uniquely the La–OH sites displays a narrower lineshape and higher ¹⁷O chemical shift. The increase in chemical shift is expected from a La metal-induced shift. , Qualitatively, this reveals that there are both Al- and La-terminated portions on the surface.

Collectively, these spectra (Figures and ) revealed the surface of LaAlO₃ to be terminated by four-, five- and six-coordinate Al, La, and hydroxyl groups that are bound to both Al and La. The relative spectral intensities of the Al–OH and La–OH environments in the ¹⁷O spectra suggest that the surface is terminated mostly by Al–OH. These spectra, however, are not quantitative and it is possible that the surface reconstructions may have been modified by the ¹⁷O enrichment procedure. Moreover, it remains unclear whether the surface is terminated by mixed LaAlO _x layers or by distinct facets of monolayers (or bilayers) of AlO _x and LaO _x (Figure ). ,

3.3. Surface Termination of LaAlO₃ Nanoparticles

The different proposed surface termination structures may be differentiated by the effective dipolar couplings between the metal sites (²⁷Al, ¹³⁹La) and the atoms from the surface hydroxyl groups (¹H, ¹⁷O). Because hydroxyls are only found at external surfaces, their dipolar couplings inherently probe the depth of the nearest Al and La layers. Furthermore, because dipolar recoupling experiments are quantitative, and can be accurately predicted in silico from a given surface model, they can in principle be applied toward the quantification of the four surface terminations (Figure ).

To this end, we measured the ¹⁷O­{²⁷Al} TRAPDOR dephasing of the surface sites as well as their ¹H­{²⁷Al} and ¹H­{¹³⁹La} S-RESPDOR dephasing. While the former may have undergone additional reconstruction during treatment, the latter probes the surface of untreated LaAlO₃ nanoparticles. The three experimental dephasing curves are shown in Figure together with numerical simulations calculated using each of the four terminations depicted in Figure . None of these calculated dephasing curves reproduced the initial rise in the experimental S-RESPDOR data, suggesting that the surface is composed of a number of distinct terminations.

6.

(Black dots) Experimental ¹⁷O­{²⁷Al} TRAPDOR, ¹H­{²⁷Al} RESPDOR, and ¹H­{¹³⁹La} RESPDOR dephasing curves. The simulated dephasing curves (solid-colored lines) correspond to simulations employing the individual surface terminations. A legend correlating the color of the simulated dephasing curves to the structural models used is provided on the left.

To find the best-fit to the experimental dephasing curves, simulated dephasing curves were calculated by employing pairwise combinations of the four surface terminations (Figures S7–S12), in relative population increments of 1%. Of the six pairwise combinations, only two yielded models with simulated dipolar dephasing within 1σ of uncertainty (Figure ), as determined by χ² analysis, with the best-fit corresponding to 67% monolayer AlO _x and 33% mixed LaAlO _x termination. The average surface populations of AlO _x and LaO _x for the models within 1σ of uncertainty were 82 ± 5% AlO _x and 18 ± 5% LaO _x , indicating that the surface is largely terminated by AlO _x .

7.

Calculated χ² values (blue curve) obtained from pairwise combinations of the surface terminations. The black line corresponds to the minimum χ² value, and the green line corresponds to 1σ of uncertainty.

Since multiple pairwise combinations yielded models within 1σ of uncertainty, an extension of this methodology was employed to consider combinations that included all four surface terminations (ternary plots, that show the edges of the quaternary phase space, are included in Figure ), yielding a total of 176,851 surface termination models. The ternary plots revealed that many models with various combinations of surface terminations fell within 1σ of uncertainty. Interestingly, despite the inclusion of multiple surface terminations, the best-fit model was still the pairwise combination of AlO _x monolayer and mixed LaAlO _x termination. The average of the relative populations of the four surface terminations for all models within 1σ uncertainty were 62 ± 10% monolayer AlO _x , 11 ± 7% monolayer LaO _x , 8 ± 6% bilayer AlO _x , and 19 ± 12% mixed LaAlO _x terminations. This average corresponds to the surface being terminated by 80 ± 4% Al and 20 ± 4% La, in agreement with the results from the pairwise combinations. However, since the uncertainties (i.e., ±6 to 12%) for monolayer AlO _x , monolayer LaO _x , bilayer AlO _x , and mixed LaAlO _x termination are smaller than their respective average, it is not possible to rule out any of them entirely given the complexity and limited resolution of the experiment. Nonetheless, the average of these models revealed that the surfaces of these LaAlO₃ nanoparticles are mostly AlO _x -terminated.

8.

Ternary plots correlating calculated χ² values to the relative amounts of each surface termination and the simulated dipolar dephasing curves for the best-fit model, corresponding to 67% monolayer AlO _x and 33% mixed LaAlO _x termination. The best-fit model is indicated by a red star, and all models within the black lines are within 1σ of uncertainty.

4. Conclusion

Nanocrystalline materials undergo intricate and often puzzling surface reconstructions that are important for guiding their catalytic, optical, or other properties. At present, there are no general methods that may reveal the full surface structural landscape present in a sample. This work has demonstrated a substantial leap toward this goal, revealing that surface-selective double- and triple-resonance SSNMR methods can be applied to estimate the proportions, of the most important surface terminations in a bulk nanocrystalline sample. Specifically, we demonstrated using LaAlO₃ nanoparticles that spectral edited triple-resonance methods could be applied to measure surface-selective heteronuclear correlation spectra, in this case separating the La–OH and Al–OH ¹⁷O­{¹H} correlation spectra from La- and Al-terminated reconstructions. Surface-enhanced ²⁷Al and ¹³⁹La SSNMR spectra further revealed these to be composed of surface four-, five-, and six-coordinate Al, and La. Surface-selective dipolar dephasing experiments probing distances between ¹H and ¹⁷O to the Al and La layers of the material were used to reveal that the surface is terminated by Al and La-terminated layers, in relative amounts of 80 ± 4% and 20 ± 4%, respectively, providing the first estimate for the relative abundances of different surface terminations in a material from experiment.

While highly promising, this approach does struggle to differentiate similar surface terminations, such as distinct AlO _x monolayer reconstructions, and it is limited to surface terminations that contain surface-bound hydroxyl groups. The highest-resolution structures are still obtained from surface electron microscopy and diffractions methods. SSNMR is nevertheless useful in disentangling competing structures measured in small surface regions and help lead to a description of the ensemble surface. We envision this approach to be valuable to understand how different surface terminations are impacted by synthetic treatment conditions. In the case of materials for which the surface has not been studied using electron microscopy, surface structure prediction codes, such as CALYPSO, MAGUS, or USPEX, may be used to generate reasonable structures as starting points as these codes have been successful in reproducing complex surface terminations. −

Raw data, simulated RESPDOR and TRAPDOR data, and structure files are available at https://zenodo.org (DOI: 10.5281/zenodo.21360645).

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.6c03759.

• Experimental and calculated NMR tensors, TEM images, EDS elemental analysis, simulated NMR spectra, and simulated RESPDOR and TRAPDOR data (PDF)

The authors declare no competing financial interest.