5.3 Pedestal Characterisation
5.3.4 Benchmark
For the discussion of the results it is important to document the properties of the different methods to characterize the edge pedestal. Several tests were performed to assess these prop-erties. First, the methods were applied to simulated data with known pedestal parameters and asymmetric profile shape. Second, real data was analysed for similar discharges where the pedestal parameters are expected to be unchanged. Further tests are presented which show comparison with individual discharges and the whole database.
5.3.4.1 Tests on Simulated Profiles
The simulated pedestals are setup with three regions of constant, but different, gradients and continuous transitions between the regions. The profile is than described by 7 parame-ters, the three gradients and position and value for the two intersections. The artificial data points are distributed normally around this curve. The standard deviation of the data points is 7% in vertical and 2 mm in radial direction. These values are chosen to resemble typical measurements at AUG. An examplary artificial pedestal profile is illustrated in Figure 5.18.
One simulation consists of 500 profiles with the same properties but randomly scattered data points. Multiple simulations are selected to test and document the influences of asymetries in the pedestal. The parameters of each pedestal - top, width, gradient - are determined with the three methods: two-line, mtanh and filter. This gives one mean value with a certain standard deviation for each simulation, method and pedestal parameter.
In Figure 5.19 the results of three tests are shown. For each pedestal parameter the relative deviation from the set value is illustrated. The set value is of no real importance since it was arbitrarily chosen to match the definition of the two-line method. Therefore, a constant offset from the set value only illustrates the differences with the two-line method. Of interest are the variations in the parameters found with one method within a given test, where one or more of the set values were fixed. In the first test, shown in the first column of Figure 5.19, the pedestal itself is unchanged and only the core gradient is varied. A temperature profile is generally more peaked than the density profile. Therefore, a core gradient in arbitrary
5.3 Pedestal Characterisation 63
units of 1−5 would correspond to a density like profile and 6−12 to a temperature or pressure like profile. The filter and two-line method are not influenced by the change of the core gradient. The mtanh method reproduces the pedestal width accurately. However, it shows a clear variation of about 20% in pedestal top and gradient, although, these parame-ters were the same in these simulations. This is likely a result of the point symmetry of the hyperbolic tangent function. Because of its symmetry the mtanh’s ability to fit asymmetric profiles is diminished. Although, the additional polynomials in the mtanh should cope with assymetries, they do not resolve the problem completely.
A second test is shown in the middle column of Figure 5.19. Again the pedestal width is kept constant but the pedestal top value and consequently the pedestal gradient is varied. Filter and two-line method yield constant width and can follow the variation in the pedestal top.
The mtanh method reproduces the pedestal parameters well unless the ratio of pedestal
gra--0.4 -0.2 0.0 0.2 0.4
1.0 6.0 11.0
deviation
core gradient -0.4
-0.2 0.0 0.2 0.4
deviation
-0.4 -0.2 0.0 0.2 0.4
deviation
0.00 0.50 1.00 1.50
2.09 2.11 2.13 2.15
pedestal parameter
Rmaj [m]
0.5 0.7 0.9 1.1
pedestal top pedestal width
pedestal top pedestal gradient 2.09 2.11 2.13 2.15
Rmaj [m]
1.0 2.0 3.0
MTANH
pedestal width [cm]
TWO-LINEFILTER
2.09 2.11 2.13 2.15 Rmaj [m]
Figure 5.19: In the top row a sketch illustrates how the pedestal is varied during each simulation. In the lower three rows the results of the pedestal simulation are shown. The relative deviation of mtanh (bottom), two-line (middle) and filter method (top) from preset properties is indicated for the pedestal width (blue, square), the pedestal top (red, circle) and the pedestal gradient (green, triangle). In the left column only the gradient inside of the pedestal top is varied, while the whole pedestal is unchanged. In the middle column pedestal top and gradient are varied, while the pedestal width is unchanged. In the right column pedestal width and gradient are varied, while the pedestal top is unchanged.
64 5. Methodology
dient and core gradients becomes small. This is visible when low pedestal top values (small pedestal gradients) are compared to large pedestal top values. The mtanh method finds a 25% variation of the pedestal width, although, the width was not varied in the parameter scan.
In the third column pedestal width and gradient were varied. Pedestal top and core gradient were set to values where all three methods showed good results before. The pedestal param-eters determined with two-line and mtanh are in good agreement with each other for widths larger than 1.5 cm. The large relative deviation for the case with 1.0 cm pedestal width is due to the finite radial resolution in the simulation. The filter method does not give useful results as stand-alone method, as expected. Therefore, the width of the kernel has to be chosen appropriately, an appropriate width can be obtained with one of the other methods.
In summary, the simulation covered a variety of different possible pedestal shapes and the two-line and mtanh methods were found to agree within 10% in the pedestal parameters for most cases. The radial scatter of the data points in the simulation was normally distributed with a standard deviation of 2 mm. This scatter is represented by the error bars of pedestal width and gradient in Figure 5.19. In this case the pedestal width cannot be determined to better than ±3 mm.
5.3.4.2 Tests on Experimental Profiles
In real measurements the conditions are not so predefined as in a simulation. In order to get useful statistics an AUG standard scenario with plasma current 1 MA, toroidal field 2.5 T, heating power 6 MW and gas puffing 1.2·1022s−1 was chosen to compare the two-line and the mtanh method. The pedestal width determined with the two-two-line method depends on the pedestal top and the separatrix position. The separatrix position has to be determined separately as described in Section 7.1. For this series of over 50 independent time slices in different discharges reproducible pedestal parameters are expected. The profiles were analysed with both methods. The mtanh model yielded a mean width ∆ne= 1.8cm with a standard deviation of 0.8 cm, a pedestal top density ne,ped = (7.6±0.9)·1019 m−3, a temperature pedestal width ∆Te= (1.9±0.5)cm and Te,ped = (0.38± 0.09) keV. The results for the two-line method were ∆ne= (1.7±0.2)cm, ne,ped = (6.9±0.5)·1019 m−3,
∆Te= (1.7±0.3)cm andTe,ped = (0.43±0.03) keV. Within the uncertainties both methods result in the same pedestal parameters. However, the two-line method shows significantly reduced scatter. In Figure 5.20 this discharge set is illustrated with a histogram. The pedestal width determined with the two-line method shows a narrower and more peaked distribution for the electron temperature (a) than for the mtanh method. Both methods yield a very similar distribution for the density pedestal width (b), however, the mtanh method produces a number of outliers with ∆ne>2.0which increase the standard variation.
The observation made with real discharges shows a similar scatter as was observed in the artificial test Section 5.3.4.1. In particular, the comparison of the scatter in the temperature and density width. When ignoring the outliers in∆newith the mtanh method, the scatter is comparable to the 10% obtained with the two-line method. For∆Tethe mtanh method yields an appreciably larger scatter of about 25%. The same observation is made in the simulation when varying the core gradient from a density like profile (1 in [a.u.]) to a temperature like profile (10 in [a.u.]) as shown in Figure 5.19 in the bottom left plot. There the scatter also increases from 10% to 20% only due to a change in the symmetry of the profile.
5.3 Pedestal Characterisation 65
fraction
∆Te [cm]
two-line mtanh
0.00 0.10 0.20 0.30
0.0 1.0 2.0 3.0 4.0
(a)
fraction
∆ne [cm]
two-line mtanh
0.00 0.10 0.20 0.30
0.0 1.0 2.0 3.0 4.0
(b)
Figure 5.20: Histogram of pedestal widths of electron temperature (a) and electron density (b) for an AUG standard scenario. The width is determined with two different methods: two-line (red) and mtanh (blue).
0.00 0.04 0.08 0.12
0.30 0.60 0.90 1.20 Te pedestal width [ψN]
βp,ped0.5 two-line
AUG DIII-D
(a) JET
0.00 0.04 0.08 0.12
0.30 0.60 0.90 1.20 Te pedestal width [ψN]
βp,ped0.5 mtanh
AUG DIII-D
(b) JET
Figure 5.21: The pedestal width in normalized poloidal flux plotted against the normalized poloidal pedestal pressureβp,ped. The pedestal parameters were determined with the two-line (a) and mtanh (b) method.
A different approach to compare the methods characterizing the pedestal region, is with a database of different discharges. In this approach the difficulty arises that physics related changes can occur at the same time as method related changes. Therefore, the scatter in the data could increase for two different reasons: the uncertainty due to the applied method or simply a wrong model was used to describe the data. A plot from Chapter 7 with the two-line method was remade with the mtanh method. In Figure 5.21 the comparison shows that in general the same trends are obtained with both methods. However, the data obtained with the two-line method (a) shows a separation of the DIII-D data into two branches, one branch which matches the AUG data and another which shows a different slope. The larger scatter, in the case with the mtanh method (b), does not allow to resolve these details in the data. The second comparison shows two regressions to the pedestal width data in Figure 5.22. The resulting regression coefficients are fairly similar, however, the scatter in the data described by the root mean square error RMSE (see Eq. (5.13)) is significantly larger in the case where the mtanh method was applied to the data (b).
As last example three very similar discharges from DIII-D were chosen. The measure-ments with the TS diagnostic show a pronounced pedestal. All methods should give optimal
66 5. Methodology
0.00 0.04 0.08 0.12
0.00 0.04 0.08 0.12
Te pedestal width [ψN]
scaling with a,〈Bp〉,Te,ped,ne,ped,κ two-line
RMSE: 16.0%
AUG DIII-D JET (a)
0.00 0.04 0.08 0.12
0.00 0.04 0.08 0.12
Te pedestal width [ψN]
scaling with a,〈Bp〉,Te,ped,ne,ped,κ mtanh
RMSE: 24.7%
AUG DIII-D JET (b)
Figure 5.22: The pedestal width in normalized poloidal flux fitted with several parameters in a log-linear regression. The pedestal parameters were determined with the two-line (a) and mtanh (b) method.
Te [keV]
normalized radius ρp DIII-D #128244,45,49
two-line mtanh 0.0
0.5 1.0 1.5 2.0 2.5
0.80 0.85 0.90 0.95 1.00 1.05
Figure 5.23: Electron temperature pedestal of three discharges with only small differences in heating power (#128244: 7.1 MW - blue, #128245: 7.6 MW - green, #128249: 8.1 MW - red). The pedestal top values obtained with the two-line method are indicated as solid line and the ones of the mtanh method as dashed line.
results. This allows to get an impression of the accuracy of the methods. Besides the heating power, the discharge settings were identical for the 3 discharges. The heating was increased in steps of 0.5 MW from 7.1 MW to 8.1 MW. In Figure 5.23 the electron temperature of these discharges is plotted at the plasma edge. The electron temperature is only increasing a little at the plasma edge, but the red case with 8.1 MW shows clearly a higher Te than the blue case with 7.1 MW. The green case with 7.6 MW lies in between the other two. The pedestal analysis with the two-line method can follow this trend nicely, as indicated with the solid lines in Figure 5.23 which mark the pedestal top value. The mtanh method does not allow to resolve this trend, for all three different heating levels the same pedestal top temperature is obtained (dashed line).
To summarize, different tests were performed to assess the quality of the results with the different methods. The results of this comparison suggest that the characterisation of the pedestal with the two-line method, without the shape regularisation imposed by the