The laser-heated probe at UWA is a specially commissioned probe design manufactured by
Bruker. It is equipped with a 15mm inner diameter horizontal saddle coil supported by
a section of silica glass tube. Below the coil, a copper tube along the axis of the probe
provides access for the laser beam. It can be sealed with a ZnSe window transparent
to infrared radiation to allow air (or other gases) to be fed into the beam tube.
The laser is a Synrad 125W CO2 laser operating
at 10600nm wavelength, and is fed into the probe via a beam tube with a 90o mirror
at the bottom of the magnet. The laser is aligned onto the sample position using a HeNe
pointer laser collinear with the main laser. The beam can be focussed or broadened using
a beam expander in the vertical beam tube. The sample itself is mounted on a sample support
machined from a BN rod. This can be a flat sample table with a small rim to keep the
sample secured while inserting the probe into the magnet or a tapered tip for aerodynamic
levitation. Samples can be self-supported powder pellets or powders enclosed in a BN
container. The probe is operated inside a 9.4T wide-bore cryomagnet. The 23Na
experiment was carried out on a similar system at CRMHT.|
Fig.1: Coil section of the laser-heated NMR probe at UWA. The vertical saddle coil and supporting glass tube are at the centre. Below the base plate, the tune and match capacitors can be seen. A copper tube extending down to the base of the probe is used to feed cooling air and the laser beam into the coil section.
Fig.2a shows a comparison of the 27Al signal obtained at room temperature in the
laser-heated probe and in a standard Bruker 4mm MAS wide-bore probe (without spinning).
The sample used in the laser probe on this occasion was a pellet of 4mm diameter and
4mm thickness. The flip angle was approximately 15 degrees in both cases, and a spectral
width of 125kHz was used.
The line shape and line width is practically identical in both cases, although the signal-to-noise
ratio of the MAS probe is better by a factor of 180. This is largely due to the wider coil
diameter, which results in a much lower fill factor and r.f. field. Nevertheless, a decent quality
27Al spectrum can be taken in less than 10 seconds. However, when the probe is used at high
temperature for an extended period, the radio-frequency circuitry begins to heat up. The resulting change
in the characteristics of the oscillating circuit causes a significant deterioration of the
signal quality (cf. Fig.2b), and it may be necessary to re-tune while at elevated
Fig.2: a) Signal-to-noise of a typical static 27Al NMR spectrum (of alumina nano-particles) at room temperature in the laser probe (grey) and in a standard (Bruker 4mm wide-bore MAS) probe (black). b) Deterioration of signal quality after 30min of heating at 30W laser output (grey) due to heating of the r.f. circuitry compared to the original spectrum shortly after turning on the laser (black).
27Al NMR - Melting of refractory ceramics.|
Fig.3: Melting of Al2O3 nanoparticles. The insert shows samples after the experiment; the melted core is clearly visible in the third sample from the left.
Heating a sample with a laser beam from one side produces steep thermal gradients. This
form of heating allows us to study the structural response of refractory materials, which
are designed specifically to cope with conditions such as thermal shock, dT/dt,
and thermal gradients, dT/dx. The nature of NMR as a bulk technique,
however, means that the signal obtained is an average of all environments in the sample.
It is therefore important to judge the fraction of a phase that can be detected in a phase
Fig.3 shows a high-temperature 27Al NMR spectrum of alumina nano-particles. During the heat treatment, the particles agglomerate and subsequently sinter to form a dense refractory. The core of the specimen melts during this process as can be seen from the photographs shown in the insert. As confirmed by MAS spectra, both the nano-particles and the product consist exclusively of octahedral aluminium sites near 0ppm (relative to acidic AlCl3 solution). On top of the broad line originating from the solid, a small narrow component can be made out near 60ppm, which corresponds to motionally averaged tetrahedral sites in molten alumina. The line width of this peak is 300-350Hz, which is slightly wider than literature data (200Hz), indicating a small impurity contribution in this nano-phase material. The integral fraction of this peak in the spectrum shown is 3%. The lower limit of detection at the signal-to-noise ratio of these spectra is about 1%. These spectra were taken with 4000 averages with a 50ms delay, giving a time resolution of 200s. Since signal-to-noise goes with the square root of the number of averages, a 2% detection limit can be achieved with a time resolution of about 8s.
29Si NMR - Time resolution study.|
Fig.4: a) Spectra of SiC taken in the high-temperature probe at room temperature (black) and at 1480oC (grey). The line near -110ppm originates from the glass coil support. b) Maximum time resolution achievable for 29Si at natural abundance in an in-situ experiment. The black spectrum was averaged over 9min, the grey one over just 90s. The insert shows the top and bottom faces of a sample after heat treatment.
27Al is a rather benign NMR nucleus as long as very distinct environments such
as tetrahedral vs. octahedral or solid vs. liquid are concerned. Its short
relaxation time and high natural abundance ensure that good statistics can be obtained
in an in-situ experiment with high time resolution. As a test case for a slow-relaxing nucleus,
we have studied the 29Si spectrum of SiC at high temperatures.|
Fig.4a shows 29Si spectra taken in the laser probe at room temperature and at 1480oC. The line near 110ppm in the cold spectrum originates from the glass tube supporting the coil. The SiC line peaks near -20ppm and covers all silicon sites in the different layers of the structure. These environments are rather similar, and MAS is needed to differentiate between them as the chemical shift of these sites ranges only from -14ppm to -25ppm. The position of the line shifts slightly towards more negative chemical shift values. From thermal expansion alone, a shift in the opposite direction --if any-- would be expected because the expansion of the lattice will reduce the electron density overall. Therefore, the polarity of the Si--C bond must be changing as the temperature rises, with electron density transferred from carbon atoms to silicon atoms.
It is remarkable that a very similar signal-to-noise ratio is obtained at both temperatures if the acquisition times are compared rather than the numbers of averages. The room temperature experiment uses 144 scans with a 5s relaxation delay (9min total acquisition time), while the hot experiment takes 5612 scans at 100ms relaxation delay (12min total acquisition time) to achieve a similar signal-to-noise ratio. The reduction in signal strength that is observed is due to the fact that the hot and cold parts of the sample have considerably different spin-lattice relaxation times. At the faster acquisition rate, cold material is saturated after just one or two pulses and does not contribute to the signal. Fig.4b explores the limit of time resolution for 29Si, again taken as representative for a nucleus with rather low natural abundance and slow room-temperature relaxation. A signal-to-noise ratio of 3:1, limited to the hot fraction of the sample, can be achieved with about 900 scans, giving a time resolution of 90s.
23Na NMR - Glass forming kinetics.||
As an example of the usefulness of laser-heated high-temperature NMR in complex materials
of industrial relevance, we have studied the reaction of sodium carbonate and quartz below, at, and
above the liquidus temperature of sodium disilicate, Na2Si2O5.
Fig.5 shows a 2D plot of the 23Na spectra obtained as a function of time. Each slice represents
a spectrum averaged over one minute (128 averages with a delay of 460ms). The third and last spectra
are shown superimposed on the plot. There is a gradual movement of the median of the line from the
sodium carbonate position at 1ppm to that of sodium metasilicate, Na2SiO3, at 6ppm.
The metasilicate peak is significantly narrower than the sodium carbonate peak of the original spectrum.
The fact that the change in line position is gradual rather than a growth of a new component on top
of the spectrum of the raw material suggests that the reaction progresses by sodium diffusion, a process
during which the quartz grains are restructured to form metasilicate. Carrying out the experiment
above the liquidus temperature of disilicate, the lowest-melting compound in the
system, changes only the rate at which the reaction occurs but does not result in the detection of a
liquid phase at any point. This indicates that sodium diffusion is sufficiently fast that the reaction
front can pass through the region of the sample which is above the liquidus temperature without the
formation of the equilibrium phase, i.e. disilicate melt.|
Fig.5: Evolution of in-situ 23Na spectra of a glass-forming batch at 855oC with time (evolving from bottom to top). Each horizontal slice is averaged over a 1min period. The shades of grey represent amplitudes. Spectra taken after 3min and after 80min, corresponding to the bottom and top slices in the two-dimensional plot in the background, are superimposed.
In order to quantify the reaction kinetics it is necessary to deconvolve the spectra. Fig.6 shows the
five components used in fitting each slice of the in-situ high-temperature experiment. The system is too
complicated to model the data with a quadrupolar line shape: Raw material and product line shape overlap
to a large extent, and the lines are partially motionally narrowed. However, we do know the line positions
of both the raw material and the product and can simulate the line shape empirically by overlapping
Gaussians representing sodium carbonate and metasilicate and a contribution common to both to cover the
quadrupolar tail of either line. It turns out
that this tail component does not vary very much
with time and therefore does not produce a considerable error. In addition, a small fixed Gaussian
contribution from the probe background is also included. From the fit results, the integral of the
two main Gaussian lines represents the ratio of product and raw material (given that the tail component
does not vary substantially) and can be taken as a reaction co-ordinate to describe the progress of the
reaction. At all temperatures, the reaction kinetics appears to be saturating after a while, which is
a more deceleratory behaviour than any of the standard models for diffusion-controlled growth predicts.
This is caused by the thermal gradient within the sample, which results in a top layer which is not
reactive on the time scale of the experiment.|
Fig.6: Deconvolution of a wide-line high-temperature spectrum. The four Gaussian components relate to the product (Na2SiO3), raw material (Na2CO3), probe background, and quadrupolar tail (peak maxima from left to right).
|Fig.7: Arrhenius plots of the spin-lattice relaxation rate vs. inverse temperature of amorphous (a), nano-crystalline (n2), and macro-crystalline (h) LixTiS2. Relaxation in these systems is facilitated by mobile Li+ ions; the slope of the curve corresponds to the activation energy of short-range ionic motion.||
Boltzmann vs. relaxation.|
Spin-lattice relaxation also compensates, for the most part, for any "in-situ penalty" resulting from the reduced Boltzmann polarisation at high temperature. The signal to be obtained from a transition between energy levels E1 and E2 is proportional to the Boltzmann polarisation
where N1,2 are the respective populations, kB is Boltzmann's constant, and T is the temperature. Therefore the signal obtained in a single scan is proportional to e-T. At the same time, the spin-lattice relaxation rate increases following a similar Arrhenius-type law,
where Ea is the activation energy for the process that facilitates relaxation and T1-1 is the spin-lattice relaxation rate. Since the relaxation delay needed to ensure that the sample is close enough to equilibrium before the next scan is proportional to the relaxation time (typically 5T1 is used), the number of scans that can be taken during a fixed period of time is also proportional to e-T, just compensating for the loss in Boltzmann polarisation. There are two limits to this mechanism, though. Fig.7 shows 7Li relaxation rate data of various forms of LiTiS2 as an Arrhenius plot. In this representation, the slope of the curve represents -Ea/kB. It is clear from the graph that the Arrhenius relationship breaks down both at very low and very high temperatures. Only if spin-lattice relaxation is an activated process, e.g. if ionic motion or the creation of phonons is the mechanism that allows the energy from the NMR transition to be dissipated, the favourable temperature dependence is obtained. While low-energy mechanisms at the cold end of the temperature scale such as localised motion processes are not relevant here, the relaxation rate maximum is likely to be reached in some laser-heated in-situ experiments. This rate maximum is caused by a resonance of the Larmor frequency of the NMR transition and the frequency of the underlying relaxation process. If, for example, the activated relaxation process is ionic motion, then the rate maximum occurs at the temperature at which the hopping frequency of the ions matches the Larmor frequency of the spin system. This typically occurs near the melting point of a material because spin-lattice relaxation and structural relaxation are closely coupled processes. It must therefore be expected that the signal quality of a high-temperature NMR experiment will deteriorate near the melting point.