Effect of coherent edge-localized mode on transition to high-performance hybrid scenarios in KSTAR

This paper deals with one of the origins and trigger mechanisms responsible for the observed performance enhancements in the hybrid scenario experiments conducted in Korea Superconducting Tokamak Advanced Research (KSTAR). The major contribution to the performance improvement comes from a broader and higher pedestal formation. The increase of fast ion pressure due to a plasma density decrease also contributes substantially to the global beta. Although the reduced core plasma volume resulting from the pedestal expansion has a negative effect on the core thermal energy, a considerable confinement improvement observed in the inner core region limits the degradation. The one significant characteristic of high-performance discharges is the presence of Coherent Edge-localized Mode (CEM) activity. CEM is triggered during the pedestal recovery phase between typical ELM crashes and has been found to be related to the increase of particle and heat transport. It appears to underlie two commonly observed phenomena in high-performance hybrid scenario discharges in KSTAR; pedestal broadening and continuous density decrease. Despite the associated transport increase, CEM activities can induce performance enhancement. With the pedestal broadening, ELM crashes become delayed and weakened, which, in turn, allows for a higher pedestal. Moreover, the density decrease directly increases fast ion pressure by extending the beam-slowing-down time. The linear gyrokinetic analysis reveals that the increase of fast ions could initiate positive feedback loops, leading to the stabilization of Ion Temperature Gradient mode in the inner core region.

Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction
Referred to alternatively as advanced inductive scenarios [1] or improved H-mode scenarios [2], hybrid scenarios [3,4] are one of the advanced operation scenarios relying on the Hmode pedestal.It characterizes a large volume of low magnetic shear in the core with a minimum safety factor, q, above or near unity.While the precise mechanism of clamping the q-profile has not been clearly identified such as flux pumping [5] or additional current sources [6][7][8], hybrid scenarios have better stability and improved confinement, due to these characteristics.For example, the absence of a q = 1 flux surface effectively suppresses well-known sawtooth instabilities that can trigger the disruptive Neoclassical Tearing Mode (NTM) [3,9], thereby permitting higher pressure operations.In addition, low magnetic shear is known for enhancing the Electromagnetic (EM) stabilization of the Ion Temperature Gradient mode (ITG) turbulence [10].Furthermore, the resulting high magnetic shear towards the outer radii increases the ITG threshold for R/L Ti [11], the normalized inverse gradient length of the ion temperature.For this reason, hybrid scenarios are prescribed as a long-pulse operation mode to test blankets in International Thermonuclear Experimental Reactor (ITER) [1,3,4,12] where higher fusion gain over the baseline scenarios is obtained at reduced plasma current, which can be utilized for the validation of high fusion gain operation for a longer duration.
Korea Superconducting Tokamak Advanced Research (KSTAR) [13,14] is addressing the establishment of highperformance, steady-state operations with various advanced operation scenarios [15][16][17][18][19][20][21][22].Among them, hybrid scenarios have been explored for years [23,24].Hybrid scenarios have been attempted by utilizing three co-direction on-axis Neutral Beam Injection (NBI) sources providing up to 5 MW as the primary heating source [25].Operating within the q 95 range of 3.5-6.5,where the sawtooth presents inherently, hybrid scenarios cover the q 95 values of the candidate scenarios in ITER [3].As a result, high-performance hybrid discharges characterizing β N up to 3.0 and a G-factor up to 0.4 in the stationary phase without sawtooth instabilities have been achieved, where G-factor is the normalized fusion gain defined as β N H 89 /q 2 95 .Including the aforementioned, the most recent work [24] focused on reporting the overview of hybrid scenario experiments operated in the 2014-2017 KSTAR campaigns.It established the target operation window for hybrid scenario experiments and defined hybrid scenario discharges in KSTAR by considering the limitation of diagnostics and analysis abilities at that time.Moreover, it introduced various experimental approaches to obtain hybrid modes in KSTAR and sorted out the observed dominant core Magnetohydrodynamics (MHD) modes.In addition, a considerable effort was made to explain the underlying physical mechanism supporting the high performance by analyzing the slow H-to hybrid mode transition period after the third NBI in a representative hybrid mode discharge.As a result, several mechanisms have been proposed, including the core-edge coupling.
Even though the previous work covered a wide range of subjects related to hybrid scenario experiments in KSTAR, issues related to high-performance transition or triggering mechanisms remain unsolved.For instance, the phenomenon of plasma density decrease with performance enhancement has been noted, which is responsible for enhancing fast ion contents.However, the reasons for the decreasing plasma density still need to be clarified.Similarly, the phenomenon of pedestal broadening observed during performance enhancement needs further exploration.The previous work primarily underscored the elevation in pedestal height attributable to improved ideal Peeling-Ballooning Mode (PBM) stability following the global beta increase.However, the slower enhancement of α max , the maximum local pressure gradient in the pedestal, of the equilibrium points compared with the expansion of the ideal PBM stability boundary, such as the change from 5.0 s to 5.3 s in figure 24(b) in [24], implies the presence of an additional physical mechanism related to pedestal broadening.Additionally, some recent hybrid scenario experiments following the aforementioned study have presented no performance enhancement and remained comparable to the conventional Hmode performance despite the same operational setting.These require studying the high-performance transition or trigger mechanism more systematically.
This paper suggests a coherent edge-localized MHD mode as one of the primary triggers for high performance in KSTAR hybrid scenarios.In several hybrid scenario experiments conducted in the 2020 and 2021 campaigns, we observed that two hybrid scenario discharges operated with the same operational setting exhibited significantly different performances in the stationary phase.The high-performance discharge featured a broader and heightened pedestal, characterized by a reduced pedestal pressure gradient and a decreased plasma density compared with the low-performance shot.Interestingly, the primary difference was the presence of the edge-localized MHD mode activities in the high-performance discharge.This mode is observed between edge localized mode (ELM) crashes within typical ELM cycles and tends to increase particle and heat transport in the edge, which are seemingly related to the aforementioned attributes observed in the highperformance discharge.It is worth noting that this trend is consistently observed across other hybrid scenario target experiments, including the discharges in the previous work [24].
Similar modes have been reported from various machines [26][27][28], such as ELM precursors, inter-ELM modes, or coherent modes (CMs).For instance, JET analyzed the ELM precursors, characterized by extended lifetimes, intermediate toroidal mode numbers, and kink-like structures [29], and also revealed that they follow ideal PBM stability [30].EAST reported an increased particle transport during the CM activities [31] and a correlation between CM activities and ELM frequency [32].ASDEX Upgrade has also characterized inter-ELM modes [33].In KSTAR, dynamics of the pre-ELM crash MHD modes have been investigated [34][35][36][37] utilizing Electron Cyclotron Emission Imaging (ECEI) [38].As will be discussed in section 3, the characteristics of the coherent edge-localized MHD mode observed in the high-performance hybrid scenario discharges in KSTAR look similar to those above.Nevertheless, the effect of the mode is unique, such that it significantly changes the pedestal profiles and triggers the transition to the high-performance phase.Since the main focus of this paper is reporting the effect of the mode on the high-performance transition rather than scrutinizing the mode itself, the identification of the mode is deferred to future work.Instead, for convenience in subsequent sections, we will refer to this mode as 'CEM', an abbreviation denoting Coherent Edge-localized Mode.
The detailed analysis will be shown in the following sections.Section 2 provides an illustrative instance where two discharges operated with the same operational settings show significantly different performances; one exemplifies highperformance discharge with CEM activities, while the other shows low performance without CEM activities.A comprehensive performance analysis follows, investigating the origins of the enhanced performance.Section 3 introduces the characteristics of CEM, discusses how to access the CEMdominant regime, and illustrates pedestal broadening and a plasma density decrease, two primary changes related to increased transport under CEM activities.In sections 4 and 5, we investigate the performance enhancement in the edge and core, which correlates with the aforementioned phenomena of the pedestal broadening and plasma density decrease.Finally, in the last section, we present a summary of our findings and concluding remarks.

Overview of the similar high-and low-performance discharges
With the same target operation scenarios, such as the same waveforms of the plasma current, external heating, and boundary shape, two discharges were selected that ended up with different plasma performances in the stationary phase.Shots 25452 and 25458 in figure 1 were operated with the same operational settings, but β N and the neutron rate, indicators of plasma performance, were different in the stationary phase.The high-performance case, shot 25452, exhibits β N = 2.8, whereas the low-performance case, shot 25458, has β N = 2.2.Furthermore, the neutron rate is nearly twice as high in the former compared to the latter case.Interestingly, plasma performances are comparable until 2.1 s, when the L to H mode transition has already been completed and conventional ELM cycles are ongoing.It suggests that another physical mechanism, besides the intended operating conditions, is required to comprehend the deviation in plasma performances after 2.1 s.
While the plasma performances differ, the plasma density and the toroidal rotation vary as well.Despite the absence of external gas fueling in both shots, the high-performance case exhibits slightly larger line-averaged density, n e , and H-alpha line intensity, H α , in the early phase, presumably due to the wall retention uncontrollable in KSTAR.However, n e is reduced during the performance enhancement period from 2.1 s to 2.8 s in the high-performance case.n e is sustained at about 25% lower value than that in the lowperformance case during the stationary phase.Meanwhile, the difference in H α between the two cases remains constant.As neither external gas fueling nor density feedback control was applied, the decrease in n e is believed to be the result of increased particle transport during the performance enhancement.The total radiation power measured by the bolometer follows n e , displaying lower radiation power during the stationary phase in the high-performance case.
Similarly, the edge toroidal rotation, V tor,e , is reduced in the high-performance case.Unfortunately, the V tor,e evolution cannot be traced due to the lack of measurement in the early phase.The decrease in V tor,e may stem from either increased toroidal momentum transport, like in the case of n e , or reduced residual stress while with a similar external momentum input.Note that residual stress is known to be the primary toroidal momentum source in the edge region in KSTAR H-mode discharges [39] and is highly reliant on the pedestal structure.Therefore, pedestal broadening in the high-performance case, as represented in section 2.2, may cause a reduction in V tor,e .In conclusion, performance enhancement observed in the high-performance case accompanies considerable changes in plasma characteristics, and they never return, which seems like a transition phenomenon.
The n = 1 MHD mode activities are displayed in the fifth row of figure 1.In both cases, m/n = 1/1 internal kink-like modes were observed until 2.4 s, followed by the disappearance of those modes.During the stationary phase, m/n = 1/1 fishbone modes prevailed solely in the high-performance case.Here, m and n are poloidal and toroidal mode numbers, respectively.These m/n = 1/1 modes are familiar in KSTAR hybrid scenario discharges [24].A notable distinction between the high-and the low-performance cases is the occurrence of MHD mode activities in the edge region.As presented in the last row of figure 1, robust MHD mode signals with toroidal mode numbers of 5-8 were observed in the edge region during every inter-ELM crash period after 2.1 s in the highperformance case.The dominant amplitude of the magnetic fluctuations originates from n = 6 or 7, persisting throughout the entire stationary phase in the high-performance case.On the contrary, in the low-performance case, the mode activities were rarely observed, with relatively low amplitude even if they are present.
As demonstrated in the following subsection, the increase in β N observed in the high-performance case is mainly due to the enhancement of pedestal pressure with the width broadened.Furthermore, the m/n = 1/1 MHD modes in the core region are located near the magnetic axis, such as ψ N ≲ 0.1, which is not associated with the change in the pedestal.This suggests the possibility of a correlation between the mode activities in the edge and performance enhancement.It may seem counter-intuitive as MHD mode activities generally increase transport and degrade confinement.Nevertheless, this is feasible when taking into account the pedestal dynamics.This is because the primary limitation of H-mode confinement pertains to energy loss caused by recurrent ELM crashes rather than pedestal transport characteristics.As already pointed out in the EPED model [40], a widely accepted model for predicting pedestal, increased transport reduces the pedestal pressure gradient by broadening the pedestal width, which delays following ELM crash and consequently leads to the formation of a higher pedestal.In experiments, improved quiescent H-mode has been achieved by forming a wide pedes-tal through increased transport provided by broad-band EM fluctuations [41,42].Therefore, this paper proposes that the MHD mode activities in the edge serve as the primary trigger for the high-performance transition in KSTAR hybrid scenarios.As mentioned in section 1, this study refers to the MHD mode in the edge as CEM, and a detailed analysis is given in the subsequent sections.

Origin of performance enhancement
Before analyzing the effect of CEM activities on performance enhancement, the origin of performance enhancement in the high-performance case presented in section 2.1 will be explored first.For the comparative analysis, 'Low' refers to 2.4 s in the low-performance case, shot 25458, to represent the low-performance phase.On the other hand, 'High: early' and 'High: stationary' denote 2.4 s and 4.05 s in the high-performance case, representing the early and stationary phases, respectively.It should be noted that 'High: early' refers to when the CEM amplitude is close to saturation while plasma density keeps decreasing and plasma performance is being enhanced.On the other hand, 'High: stationary' represents a completely saturated phase in the high-performance case, considering the current diffusion time in the core region, which is approximately 1 s [24].
Figures 2(a)-(d) display the measured kinetic profiles for the three phases.Thomson scattering diagnostics [44] provided measurements for electron density, n e , and electron temperature, T e , while Charge Exchange Spectroscopy (CES) [45] measured ion temperature, T i , and toroidal rotation, V tor .Because the temporal and spatial resolution of the Thomson scattering diagnostic is not adequate for tracking the pedestal evolution during ELM cycles, the pedestal widths of n e and T e are presumed to be the same as that of T i .Additionally, n e obtained from five Two-Color Interferometers [46] with five distinct line integration paths was utilized to constrain the n e profile fitting, thereby compensating for the uncertainty present in the density profile measurement.T e = T i was also assumed in the pedestal region due to considerable measurement uncertainty in the edge T e .Using kinetic-EFIT [47], equilibria were reconstructed to fit profiles in the normalized poloidal flux coordinates, ψ N , and to obtain safety factor profiles, q.Kinetic-EFIT employs measured kinetic and magnetic pitch angle profiles as supplementary constraints in the equilibrium reconstruction based on external magnetic measurements.Unfortunately, Motional Stark Effect diagnostics [48] were not available in these experiments for measuring magnetic pitch angle profiles.Instead, the synthetic magnetic pitch angles calculated by the equilibrium solver, CHEASE [49], and the external current drive modules were used as a constraint for the kinetic-EFIT reconstruction.Figure 2(e) represents the thermal, P th , and fast ion pressure, P f , profiles utilized to reconstruct equilibrium, whilst figure 2(f ) illustrates the safety factor profiles, q, obtained from the resultant equilibria.P th was calculated using P th = T e n e + T i n i , where n i is the thermal ion density calculated via the quasi-neutrality condition with the assumption of fully stripped carbon impurities and an effective ion charge of 1.2 for all radii.Since P f cannot be measured in KSTAR, NUBEAM [50] has calculated it by assuming zero anomalous fast ion loss considering no signs of fast ion confinement degradation between phases.It is noteworthy that the total stored energy, which was calculated by the volume integration of 3/2 (P th + P f ), has been compared to that provided by EFIT in each phase.The difference between the two is less than 1%, which confirms the validity of the assumptions made in P th and P f .The most notable observation in the kinetic profiles is the variation in the pedestal width for each phase.It is evident that the pedestal width in T i broadens as the performance improves.Additionally, P th reveals an increase in the pedestal height along with a decrease in the gradient.The difference in the P th pedestal is much more substantial between 'Low' and 'High: early' than between the two 'High' phases.Notably, the gradient of P th of ψ N = 0.8 − 0.9 is reduced between 'Low' and 'High: early', which restricts the upshift of the core P th through stiffness.Another intriguing observation is a significant increase of T i in the inner core region in 'High: stationary' compared to 'High: early'.A similar trend is found in the core T e , albeit less pronounced.Consequently, a peaked P th appears within the inner core region in 'High: stationary' when compared to the other two phases.Meanwhile, P f in figure 2(e) steadily rises during the performance improvement.In both 'High' phases, V tor significantly decreases in the edge region compared to 'Low', while the core gradient increases.The alterations in both the width and gradient of the V tor pedestal correspond with P th .
In figure 2(f ), the central q, q 0 , in all phases is above unity, indicating the absence of sawtooth instabilities in both high-and low-performance cases, thus confirming the hybrid scenarios.In 'High: early', the centrally peaked Ohmic current decreases due to increased neutral beam-driven current and edge bootstrap current, leading to a rise in q 0 compared to 'Low'.On the other hand, in 'High: stationary', the core bootstrap current increases significantly due to the enhanced pressure gradient and lower density.This leads to a decrease in q 0 and a formation of lower magnetic shear region in ψ N < 0.2.From 'Low' to 'High: stationary', q and the magnetic shear continue to decrease in the reduced magnetic shear region of ψ N > 0.9 because of an increase in the edge bootstrap current.It is observed that the fraction of the bootstrap current is approximately 0.43, and the total noninductive current fraction is approximately 0.64 in 'High: stationary'.
Based on P th and P f as presented in figure 2(e), figure 3(a) depicts the development of each energy component between each phase, namely the core thermal energy, the edge thermal energy, and the fast ion energy.As illustrated in figure 3(b), the edge thermal energy is defined as the thermal energy of the pedestal structure, while the core thermal energy makes up the remaining thermal energy.The fast ion energy is computed by volume integration of P f in the radial direction.
Despite the decreased pressure gradient of the pedestal as a result of broadening, increasing pedestal height enhances edge thermal energy by 51.1 kJ in 'High: early' compared to 'Low'.Meanwhile, the core thermal energy diminishes by 31.4 kJ while maintaining a similar P th shape.This is primarily due to the decrease in the core volume resulting from edge expansion and partly to the reduced P th gradient in the outer core region of ψ N = 0.8 − 0.9.In the case of P f , it experiences global elevation, resulting in an increase in fast ion energy by 15.2 kJ.In 'High: stationary', the pedestal width and height increase while the pedestal pressure gradient decreases, leading to a rise in edge thermal energy by 15.9 kJ compared to 'High: early'.Although the core volume is further reduced, the considerable increase of P th in the inner region of ψ N < 0.4 compensates for it.P f also rises by 17.3 kJ, a similar amount as before.In summary, the performance improvement from 'Low' to 'High: stationary' is due to the increased edge thermal energy and fast ion energy, enhancing 67.0 kJ and 32.5 kJ, respectively.While the core thermal energy experiences some loss as a consequence of edge expansion, primarily from 'Low' to 'High: early', the decrease is restricted by the considerable improvement in the inner core region from 'High: early' to 'High: stationary'.
Figure 4 presents the effective heat diffusivity for each phase, calculated by power balance analysis.From 'Low' to 'High: early', effective ion and electron heat diffusivities, χ eff i and χ eff e , increase near ψ N = 0.85, where the gradient of P th is reduced as presented in figure 2(e).This indicates that the increased ion and electron heat transport may be the reason for the reduced gradient of P th .When comparing 'High: early' and 'High: stationary', it is evident that there is a reduction in χ eff i and χ eff e in the inner core region.This finding suggests that the improvement in P th for ψ N < 0.4 originates from the reduced ion and electron heat transport.It is noteworthy that the decrease in χ eff i is more substantial than that in χ eff e , which is consistent with the core T i increased more than the core T e as depicted in figures 2(b) and (c).Meanwhile, the radial expansion of the low χ eff i and χ eff e region near ψ N = 0.95 represents pedestal broadening.In summary, the performance enhancement observed can be categorized into three main parts.Firstly, there is an enhancement of edge pressure from 'Low' to 'High: early', where 'Low' and 'High: early' exhibit rare and intense CEM activities, respectively.Secondly, there is an increase in the inner core pressure from 'High: early' to 'High: stationary', which is believed to be a result of improved thermal confinement.Lately, there is a consecutive increment in the fast ion pressure from 'Low' to 'High: stationary'.Therefore, the subsequent sections concentrate on the enhancement of the edge pressure due to the presence of CEM activities (section 4), the potential cause of the improvement in core confinement (section 5.2), and the origin of the fast ion enhancement (section 5.1).

CEM
Before analyzing the effect of CEM on edge pressure enhancement, this section introduces the characteristics of CEM, the way to access the CEM dominant regime, and the increased transport under CEM activities.

Figures 5(a) and (b) show the spectrograms of the normalized beam emission intensity fluctuation measured by Beam
Emission Spectroscopy (BES) [51], which represents the normalized electron density fluctuation near the separatrix, and the magnetic field fluctuation measured by the Mirnov coil in the high-performance case.In the example case of figure 5, CEM was found to be a coherent MHD mode with an intermediate toroidal mode number such as 6 or 7.The frequency of CEM measured in the lab frame is comparable to the toroidal rotation frequency in the edge region, considering its toroidal mode number, which implies that the mode frequency is much lower than the rotation frequency.It is worth noting that the change in the frequency of n = 6 CEMs detected before and after 2.16 s indicates the decrease of V tor,e after 2.16 s, which was presumed in section 2.1.CEM was triggered during the inter-ELM crash period, specifically in the middle of the pedestal recovery phase, and was stabilized after the following ELM crash.Sometimes, it was intermittently stabilized before the ELM crash and retriggered, such as around 2.3 s and 2.36 s in figure 5.During intermittent stabilization, a small H α peak was observed.After CEM was triggered, it grew slowly or almost saturated in amplitude, which characterizes a long lifetime of up to several tens of milliseconds and sometimes accounts for about 90% of the inter-ELM crash period.It should be noted that the n = 1 coherent mode is not CEM but an internal kink-like mode located in the inner core region with a poloidal mode number of 1.
Figure 6 presents the amplitudes of the normalized fluctuation measured by BES and ECEI with the T i profile.BES represents the electron density fluctuation, covering the lower part of the pedestal up to the Scrape-Off Layer (SOL) due to the strong beam attenuation in the upper pedestal, as commented in [52].On the other hand, ECEI measures the electron temperature fluctuation within the Last Closed Flux Surface (LCFS).ECEI reveals that CEM is localized in the edge region with a peak amplitude just inside the T i pedestal top.BES shows consistent results in the steep pedestal region, except for the beam attenuation region.Moreover, BES also reveals a finite CEM amplitude in SOL, which was identified by 2D BES imaging (not shown here) as filamentary structures connected to the mode structure inside the LCFS.The large error bars in the BES channels close to the limiter are due to the reduced beam emission intensity.Additionally, there is no observable tearing parity in the valid region of ECEI and BES, indicating that the fluctuations cannot be considered a tearing mode.
As mentioned in section 1, CEM is similar to the pre-ELM crash MHD modes [34][35][36][37] observed in KSTAR ELMy Hmode discharges.However, CEM can be distinguished from the pre-ELM crash MHD modes by its much weaker activities.Specifically, CEM grows relatively slowly or remains almost saturated, allowing it to have a much longer lifetime of over ten milliseconds, an order of magnitude longer than typical pre-ELM crash MHD modes.In addition, the peak of H α during intermittent stabilization is markedly lower than that observed during typical ELM crashes.Moreover, it was found that the T e pedestal top hardly changed during the intermittent stabilization of CEM, while it collapsed significantly during typical ELM crashes.Those differences are apparent in figure 9, where the intermittent stabilization of CEM is at 2.698, 2.727, and 2.759 s, and the typical ELM crash was at 2.706, 2.732, and 2.772 s.Therefore, CEM is assumed to be a weakened or saturated PBM-like mode, although a systematic stability analysis has not yet been performed.In the meantime, CEM occasionally grows until the following typical ELM crash, as shown in figure 10, but whether it can trigger a typical ELM crash or not is still unknown.

Accessing CEM dominant regime
As introduced in section 2.1, the only difference between 25452 and 25458, where the CEM activities are significantly different, is n e and H α .Vigorous CEM activities were obtained in the higher n e and H α case, whereas not in the lower n e and H α case, suggesting that n e and H α are key parameters for CEM activities.
The linear MHD stability analysis supports the possible correlation between the plasma density and the CEM onset.In figure 7(a), 28029 is compared to 28025, another experimental comparison to investigate the CEM activities depending on the external gas fueling rate and to check the reproducibility of the CEM-dominated plasma as in 25452.The experiment was conducted with the same I P and toroidal field as 25452, albeit with 1 MW lower heating power.A lower heating power results in slower ELM cycles, which is necessary for observing ion temperature profile evolution between ELM cycles.External gas fueling was only applied in 28029, leading to larger n e and H α compared to 28205 after 2.05 s.An intense CEM activity was triggered during the ELM cycle that began at 2.31 s in 28029 but not in 28025.
Figures 7(b) and (c) show the well-known j − α diagrams analyzed at 2.320 s and 2.352 s, respectively.These specified time slices are shown as t 0 and t 1 in figure 7(a) and represent the initial phase of the pedestal recovery with no CEM in both discharges and the time when CEM is just triggered in 28029 but still absent in 28025, respectively.The linear analysis was conducted using MISHKA1 [53], an ideal MHD stability code, assuming that the CEM onset follows the ideal PBM stability, such as the pre-ELM MHD oscillation discussed in section 3.1.For the analysis, the kinetic-EFIT equilibria were used, excluding toroidal rotation.At t 0 shown in figure 7(b), the equilibria of both discharges are in a linearly stable regime for two commonly used marginal stability criteria: γ max /ω * i = 0.25, which accounts for a diamagnetic stabilization effect, and γ max /ω A = 0.03, which does not.On the other hand, at t 1 in figure 7(c), the equilibrium of 28029 is unstable for both stability criteria, while the equilibrium of 28025 is unstable for γ max /ω A = 0.03 but stable for γ max /ω * i = 0.25.The results are consistent with the experimentally observed CEM onset when considering γ max /ω * i = 0.25, as proposed in the previous work [24].The discrepancy in the stability at t 1 is due to the effectiveness of the diamagnetic stabilization, which is inversely proportional to the plasma density.The effect of the diamagnetic stabilization on the linear stability can be estimated by comparing the gap between the two stability boundaries, which reveals that the diamagnetic stabilization effect is weaker in 28029 compared to 28025.The linear analysis suggests that CEM could be triggered in 28029 due to a higher plasma density, although 28029 has a similar pedestal pressure gradient to 28025.
In addition to the effect on the linear MHD stability, increasing n e and H α also reinforces the CEM activities.Figure 8 shows another comparative experiment, 26417 vs. 26416, with different external gas fueling rates.These discharges were operated at 50 kA lower I P than 25452 and without Electron Cyclotron Heating, which removes m/n = 1/1 activities in the central region.26 417 and 26 416 have the same n e and H α until 2.6 s, but they start to diverge after 2.6 s due to different rates of external gas fueling, reproducing the situation similar to 25452 and 25458 in section 2.1.The third row in figure 8 illustrates that both discharges have CEMs after 3.0 s, but the amplitudes of CEMs differ significantly.The high fueling discharge, 26417, has CEM activities with an amplitude more than two times larger than that of the low fueling discharge, 26416.Notably, the last row represents that the performance enhancement and the edge rotation decrease as in the high-performance discharge in section 2.1 were reproduced only in 26417, suggesting that reinforcing CEM activities by increasing n e and H α is required to access high performance.
The correlation between n e and H α and their effect on CEM activities is not fully understood at present due to measurement limitations and a lack of nonlinear analysis tools in KSTAR.

Increased transport under CEM activities
CEM activities tend to increase plasma transport.Figure 9 shows the CEM fluctuation amplitude measured by BES and Mirnov coil, H α of the lower divertor region, and electron cyclotron emission (ECE) intensity, I ECE , near the pedestal top.Between ELM crashes at 2.675, 2.706, 2.732, and 2.772 s, growing CEM amplitudes were observed in both BES and the Mirnov coil.Interestingly, the H α increased with the CEM amplitude, and the I ECE recovery slowsed down after the CEM onset.Since the H α is roughly proportional to the number of particles in the lower divertor region, one can assume that particle transport increases during CEM activities.Similarly, because I ECE is proportional to the pedestal top electron temperature, the slowing down of the I ECE recovery after the CEM onset can be interpreted as the increased edge electron heat transport during the CEM activities.The correlation between the CEM activities and the changes in H α and I ECE recovery was more apparent after 2.759 s, the intermittent stabilization of CEM; H α decreased, and I ECE recovery was accelerated.The pedestal ion temperature and the toroidal rotation also show a similar trend.Figure 10 illustrates the edge ion temperature and toroidal rotation in 28030, accompanied by the CEM fluctuation amplitude measured by BES.28030 is the same scenario as 28029 shown in section 3.2 with nearly identical n e and H α evolutions.The recovery of edge ion temperature decelerated after the CEM onset at 2.35 s, compared to the first three ELM cycles without CEM.Toroidal rotation almost saturated after the CEM onset, implying that the toroidal rotation is more susceptible to CEM activity.
Even though the location of the increased particle transport is not definite due to the limitations of the density profile measurement, the particle loss under CEM activities seems obvious.On the other hand, the evolution of the edge ion and electron temperatures reveals that the ion and electron heat transport is increased in the edge where the finite CEM amplitude is observed.In addition, it is believed that CEM activities within the pedestal top at R ∼ = 2.15m, as shown in figure 6, are responsible for the increase in χ eff i and χ eff e and the reduction in the gradient of P th in the outer core region in both 'High' phases, as discussed in section 2.2.Meanwhile, whether or not the toroidal momentum transport is increased under CEM activities is not deterministic.For example, the edge toroidal rotation could be reduced due to residual stress, as discussed in section 2.1, or neoclassical toroidal viscosity [55,56] induced by CEM.Future studies are needed to identify the dominant influence.
Accompanied by the increased edge transport discussed above, the pedestal width tends to broaden with decreasing pedestal pressure gradient under CEM activity.Figures 11(a) and (b) show T i and n e profiles near the edge at 2.8 s in 28025 and 28030, representing the pedestal profiles built up in the absence and the presence of CEM activity, respectively.Figure 11(c) displays the pressure, P, obtained by multiplying T i and n e and its gradient.The simple multiplication of T i and n e for P is to consider only the highly reliable pedestal measurements, CES and BES.Even if the reconstruction of n e utilizing BES is limited to a few time slices predefined by a unique on/off timing for each NBI source, it has the advantage of a much higher temporal and spatial resolution than the measurement by the Thomson scattering diagnostic in KSTAR.β N is 1.8 in 28025 and 2.13 in 28030.The difference in n e is due to the different external gas fueling rates, as described in the paragraphs related to figures 7 and 10.
In contrast to 28025, 28030 exhibits a larger pedestal width in both T i and n e .The pedestal widths of T i and n e are comparable to each other in both cases, supporting the assumption made in the fitted profiles in section 2.1.Specifically, the pedestal width of 28030 is almost twice that of 28025.Additionally, the gradient of the pedestal T i is lower in 28030, while the gradient of the pedestal n e is similar in both discharges.Consequently, 28030 has a reduced maximum gradient of P in the pedestal despite the enhanced pedestal top value of P. Since the external heating power is the same, the difference in the gradient of P implies changes in the transport characteristics in the pedestal.Indeed, the reduced gradient of T i for ψ N = 0.95 − 1.00, where n e is similar, implies larger ion heat transport in 28030, assuming diffusive transport.On the other hand, particle transport cannot be determined due to the expected considerable difference in the particle source and the lack of particle source modeling in KSTAR.Unfortunately, the T e pedestal difference was not examined because of diagnostic limitations.However, considering the similar evolution of the pedestal top T e and T i in the inter-ELM crash period under CEM activities shown in figures 9 and 10, the difference in the T e pedestal depending on the CEM activities is expected to be similar to T i .
For statistical purposes, figure 12 compares the measured and the predicted pedestal top and width, depending on the presence of CEM activities during the pedestal build-up.The measured pedestal parameters are from the T i pedestal selected just before the following ELM crash, and the predicted ones are from the well-known EPED model [40].The proportionality constant G, which determines transport characteristics, was set as 0.076 for all cases in the model [40].Due to the limitations of Thomson diagnostics in the edge, the density of the pedestal top was assumed to be equal to 1.2 times the line averaged density measured by the outermost interferometer, where the factor 1.2 considers that the outermost line averaged density tends to be about 20% lower than the density of the pedestal top.Additionally, the separatrix density and temperature were assumed to be 5 × 10 18 m −3 and 0.1 keV, respectively.The fixed separatrix density reflects the similar separatrix density values obtained under different external fueling shown in figure 11(b).Measured pedestals were parameterized by profile fitting utilizing the hyperbolic tangent function model [43].Input equilibria were computed by CHEASE, and linear MHD stability was analyzed by MISHKA.In the following paragraph, 'without CEM' and 'with CEM' refer to cases without and with CEM activities, respectively.
For 'without CEM', which is considered as typical Hmode pedestals, figures 12(a) and (b) show that the EPED model reproduced the measured pedestal top and width, which supports the assumptions used in the model.On the other hand,  for 'with CEM', the model reproduced the pedestal top but failed to reproduce the width.The measured width of 'with CEM' tends to be larger than the prediction.As the predicted and the measured pedestal tops are comparable, the measured width being larger than the prediction indicates a lower pedestal pressure gradient in the measurement.The model's height and width are bound by the proportionality constant G, which represents transport properties.Therefore, the lower pedestal pressure gradient of 'with CEM' implies more significant transport than what is assumed in the model.It is consistent with the discussions on the transport characteristics and the pedestal structures under CEM activities above.
Meanwhile, a commonly observed trend during the CEM dominant phase is a decrease in density, implying another phenomenon related to increased transport under CEM activities in addition to the above observations.Figure 13 compares the n e evolution in 25452, the high-performance case with CEM activities, and 25458, the low-performance case with almost no CEM activities.The n e evolution was reconstructed using five line-averaged density measurements, as introduced in figure 2(a).In 25452, n e tends to decrease at all radial locations after the first CEM onset at about 2.1 s.However, in 25458, n e starts smaller than in 25452 but steadily increases, resulting in a roughly 40% larger stationary electron density near the magnetic axis.As previously discussed in section 2.1, both discharges were operated under the same operational setting and without external fueling.Therefore, the gradual decrease in n e in the high-performance case is likely due to the increased particle transport caused by CEM activities.While it is unclear why the density decreases slowly and takes about a second to reach the stationary level, one possibility is the limited duration of increased particle transport.Because CEM activity is in the middle of the pedestal recovery phase and is sometimes intermittently stabilized, the global decrease in density could be slow, different from the rapid decrease in density usually observed in the application of resonant magnetic perturbation [57].The significant drops at 2.05 s and 2.23 s in the low-performance case correspond to the large oscillation after the initial ELM crash.

Effect of CEM on edge confinement enhancement
In the last paragraph of section 2.1, CEM activities are suggested as a primary reason for the pedestal enhancement.Figure 14 presents parameter evolutions during several ELM cycles in 26417, a high-performance hybrid scenario discharge introduced in section 3.2.Prior to 3.078 s, parameters represent the stationary phase of a typical ELMy H-mode discharge, where the repetitive pedestal collapse and the recovery kept the time-averaged total stored energy almost constant.In contrast, the subsequent ELM cycle exhibited an apparent increase in the total stored energy, with the first CEM growth starting at 3.11 s.Since no additional external heating was applied and the internal inductance or global beta evolution was already completed near 2.6 s, the abrupt performance enhancement is believed to be related to the CEM growth.
Interestingly, the performance improvement is achieved with a delay of the following ELM crash.The inter-ELM crash period lasted approximately 70 ms during the phase without CEM activities but extended to about 125 ms with the initial CEM growth.In addition, during the initial CEM growth, the total stored energy continued to increase until the following ELM crash.Generally, the lengthening of the inter-ELM crash period or the delay of the ELM crash does not necessarily imply the extension of the pedestal recovery phase.In many tokamaks, there is a saturation phase after the recovery phase where the pedestal pressure remains constant until the ELM crash [28,[58][59][60].However, in KSTAR experimental conditions, pedestal evolution in typical ELMy H-mode discharges usually does not experience the saturation phase, and the ELM crash occurs in the middle of the pedestal recovery, e.g. the ELM cycles before 2.3 s in figure 10 or before 3.0 s in figure 14.Moreover, the pedestal height continues to increase slowly despite the CEM activities, as shown by evolutions of the pedestal top T e and T i in figures 9 and 10.Hence, the delay of the subsequent ELM crash during CEM activities could potentially lead to performance enhancement in KSTAR.
Another important observation is that the performance gained during the extended recovery phase is not completely lost after the following ELM crash.For example, figures 14(b) and (f ) show that the energy loss due to the ELM crash at 3.2 s was much smaller than the energy gain prior to the ELM crash, resulting in performance enhancement.In other words, the ELM crash becomes weak in terms of the ratio between the lost and recovered energy.It implies that a finite fraction of the enhanced energy is still in the pedestal after the ELM crash.This is supported by the fact that, in the subsequent ELM cycles, CEM is triggered much earlier, and the ELM cycles tend to be more frequent.For instance, compared to the first CEM onset occurring about 0.032 s after the previous ELM crash at 3.078 s, the CEM onset in the subsequent ELM cycles was less than 0.01 s following the ELM crash.
In experiments, performance enhancement due to the delay of ELM crash occasionally happens, such as a temporary ELM-free phase right after the L-H transition or after injecting excessive external heating in a short time.However, the following large ELM crash degrades the enhanced performance and returns it to the typical H-mode performance.Therefore, it is believed that the weakened ELM crash under CEM activities is a distinctive feature and essential for the sustainability of the enhanced performance.
Figure 15 shows the evolution of the pedestal width, top pressure, and height-to-width ratio, evaluated for several ELM cycles covering the phases without and with CEM activities in 28030 introduced in section 3.3.The pedestal profiles were fitted in the same manner as the profiles in figure 2. In figure 15, ELM cycles are divided into three phases: a typical ELMy H-mode phase without CEM (A), a transiently enhanced performance phase with CEM (B), and a  permanently enhanced performance phase with CEM (C).The total stored energy is almost maintained during (A), while it is significantly enhanced in (B) and (C) with the appearance of CEM.Interestingly, in (B), the enhanced performance is not maintained and is completely lost by the subsequent multiple ELM crashes.In (C), on the other hand, small and large amounts of performance improvement are obtained around 2.70 s and 2.85 s, and the improvement is not lost after the subsequent ELM crashes.Note that the abrupt drops in the total stored energy around 2.35 s and 2.8 s were due to turning off the NBIs for the CES measurement.
For (A), the pedestal parameters stay around the relatively narrow pedestal width, low pedestal top pressure, and high height-to-width ratio, representing a typical ELMy H-mode pedestal evolution.On the other hand, for (B) and (C), the pedestal evolves to an extended pedestal width, higher pedestal top pressure, and a reduced height-to-width ratio.Especially for (B), the delay of the ELM crash is so excessive that the enormous maximum pedestal width and top pressure are obtained.Since the increase in the pedestal top pressure dominantly contributes to the edge thermal energy, the total stored energy is enhanced in (B) and (C) despite the expansion of the pedestal width and the reduction of the height-to-width ratio.It is noteworthy that the T i pedestal profile evolution in (B), which is not shown here, reveals that only the pedestal top increases at almost similar pedestal width before the CEM onset.Then, after the CEM onset, the pedestal width starts to expand, and the increase of the pedestal top slows down, as shown in figure 10.In addition, the decrease in the pedestal height-to-width ratio, which indicates the reduction in the space-averaged pedestal pressure gradient, implies an increase in edge transport, which is consistent with the discussions in the previous sections.
The primary difference between (B) and (C) is the sustainability of the enhanced edge thermal energy.In (B), after the subsequent multiple ELM crashes, the excessively increased pedestal width and top pressure return to the initial levels, similar to the minimum levels in (A).On the other hand, in (C), the pedestal width and the top pressure after the following ELM crash remain around the levels larger than the minimum values in (A) and (B).This means that in (C), a finite portion of the enhanced energy is still preserved in the pedestal even after the ELM crash, which represents a weak ELM crash under CEM activities, as discussed above.The interesting point is that the subsequent multiple ELM crashes, not a single strong ELM crash, are responsible for the drop in total stored energy in (B).For example, in the first ELM cycle in (B), the total stored energy is still larger than the minimum value in (A) after two consecutive ELM crashes, but it further decreases after the third ELM crash.In other words, a single ELM crash is weakened in (B) as well as in (C), which is consistent with the feature of a weakened ELM crash following the performance enhancement with CEM activities.The reason why the following ELM crash becomes weak is not clear, but the reduced pedestal pressure gradient under CEM activities may be responsible [61,62].
Meanwhile, the reason for the multiple ELM crashes in (B) needs to be clarified.One possibility is a large overshooting in the Plasma Control System [63,64].The increase in total stored energy causes the plasma to move outward, and the Poloidal Field (PF) coils try to push it back inward by increasing the coil current to maintain its target position.In this situation, if the plasma energy is abruptly lost due to the following ELM crash, the plasma rapidly moves inward.On the other hand, the PF coils react relatively slowly due to the superconducting feature, which could lead to overshooting.Indeed, after the ELM crash in (B), the current in the PF coils responsible for the radial position control overshoots, and the plasma moves excessively inward with its volume shrinking.Consequently, additional ELM crashes may occur because the MHD stability in the pedestal, which is responsible for ELM crashes, is sensitive to plasma position and shape [65][66][67][68][69].
The above suggests the possibility of optimizing CEM activities to achieve sustainable performance enhancement.If an ELM crash is delayed too long for any reason, the subsequent multiple ELM crashes may degrade all the enhanced performance.In a dedicated experiment, scanning density by increasing external D 2 gas fueling triggered CEM, and at high density, a delay of ELM crash became significant.This suggests a possible dependence of CEM appearance and activity on collisionality, as discussed in section 3.2.Still, there are some exceptional cases where CEM is triggered in one but not in the other despite similar plasma densities.Therefore, further analysis with improved pedestal diagnostics is necessary to optimize CEM activities and will be performed as future work.

Effect of CEM on core performance enhancement
As mentioned in section 3.3, decreasing plasma density in the CEM dominant phase is a commonly observed phenomenon in the KSTAR hybrid scenario discharges.In the example case in section 2, the high-performance case with CEM activities has about 25% lower line-averaged density in the stationary phase than that of the low-performance case without CEM.Accordingly, the fast ion pressure and thermal energy transport are expected to change depending on the plasma density.

Fast ion pressure enhancement
Since the on-axis co-current NBI system is the primary heating source providing up to 5 MW in KSTAR, a considerable amount of beam ions are usually present in the plasma core.Even in the hybrid scenario discharges, although lower than in the Fast-Ion-Regulated Enhancement mode [22] due to the higher density, fast ions still account for about 20% of the total stored energy, as shown in figure 3(a).Therefore, changes in beam or fast ion characteristics can substantially affect the total stored energy or global beta.
Under the condition where the fast ion loss is negligible, the fast ion content tends to increase with decreasing plasma density.This is due to the extension of the beam slowing down time.As the plasma density decreases, the thermalization of the high-energy beam ions is delayed due to reduced collisions between the beam ions and background particles.This means that the beam ions stay in the plasma longer, resulting in more beam ions in the plasma.
Figure 16 compares the enhancement in P f with the evolution of T 3/2 e /n e for the cases analyzed in section 2, 'Low', 'High: early', and 'High: stationary'.T 3/2 e /n e is a parameter proportional to the beam slowing down time [70,71].From 'Low' to 'High: early' to 'High: stationary', P f increases with T 3/2 e /n e , indicating that the primary change in P f is due to the extended beam slowing down time.In addition, although not shown here, the NUBEAM analysis also showed that the primary change in P f is due to the increase in the beam ion density rather than the beam ion energy, which is also consistent with the expected result of extending the beam slowing down time.Since the decrease in n e is accompanied by the increase in T e , the density decrease during CEM activities considerably enhances the fast ion energy and, thus, the total energy.Notably, the ratio of beam power loss to beam heating power increases almost exponentially with the decrease in plasma density.However, during the P f enhancement of the example case, the ratio changes only from 4.5% to 6.6%, confirming that the beam power loss is not significant during the density decrease in KSTAR hybrid scenario discharges.

Core thermal confinement enhancement
As analyzed in section 2.2, a significant improvement in the core thermal confinement was observed for ψ N < 0.4 in 'High: stationary' compared with 'High: early'.Since the entire discharge has no core MHD modes causing significant confinement loss, such as NTM, the improvement is assumed to originate from the mitigated microturbulence activities.Note that the m/n = 1/1 internal kink-like mode in 'High: early' and the m/n = 1/1 fishbone mode in 'High: stationary' are observed for ψ N < 0.1, which is not relevant for the changes in ψ N < 0.4.
For the dominant microturbulence and stabilization effects, the linear analysis was performed using the GKW code [72] at ψ N = 0.3 for 'High: early' and 'High: stationary'.The kinetic profiles introduced in section 2.2 were used as inputs.Miller geometry [73] was used to approximate the equilibria reconstructed by kinetic-EFIT.The kinetic electron was considered, and impurities were not included.Fast ions were assumed to be a separate deuterium ion species with excessively high temperatures and low densities, which were obtained from the NUBEAM calculations.The densities and density gradients of the main ions were adjusted to satisfy the quasi-neutrality condition.All EM effects were turned on to check the finite EM stabilization effects.Collision and rotation were assumed to be ignored.
Figure 17 shows the linear growth rate, γ, and the normalized real frequency, ω N , depending on the normalized poloidal wavelength, k θ ρ i , for 'High: early' and 'High: stationary', respectively.The most unstable mode is found at k θ ρ i = 0.5 and 0.4 in 'High: early' and 'High: stationary', respectively, and both propagate in the ion diamagnetic direction.Therefore, ITG is expected to be dominant in both cases.In addition, the maximum γ is lower in 'High: stationary' than in 'High: early', implying that the linear stabilization effects have been further enhanced in 'High: stationary' sufficient to compensate for the destabilization of the increased T i gradient.
To find the dominant stabilization effects in 'High: stationary', figure 18 compares the normalized linear growth rates, γ N , calculated for the cases where various local parameters are changed from 'High: early' to 'High: stationary'.Here, stabilization and destabilization mean a decrease and an increase in γ N .'High: early' in figure 18 denotes γ N calculated for 'High: early'.As a reference case to compare with the others, 'Ref.' is obtained by replacing only the R/L Ti value in the input set for 'High: early' with the value in 'High: stationary'; R/L Ti is changed from 3.86 (the value in 'High: early') to 5.48 (the value in 'High: stationary').R/L Ti is the inverse of the ITG length, 1/L Ti ≡ ∇T i /T i , multiplied by the major radius, R.This artificial case is set as a reference to represent the destabilization caused by the increase in R/L Ti in 'High: stationary' because R/L Ti is a critical parameter that determines ITG stability and is also the most significant difference in the core between the two 'High' phases.The increased γ N in 'Ref.' compared with 'High: early' for all k θ ρ i confirms the ITG dominant regime, where γ increases with R/L Ti .
The others indicate cases where, in addition to R/L Ti , another input parameter in 'High: early' is replaced by the value in 'High: stationary'.In other words, based on the input parameters for 'Ref.',where only R/L Ti is from 'High: stationary' and the other input parameters from 'High: early', another local parameter value is substituted by the value of 'High: stationary' in each case.This makes it possible to study the effect of each local parameter change between the two 'High' phases under the increased R/L Ti .For example, 'β' indicates the case where, in the input of 'Ref.', the normalized local pressure value, β, is replaced by the 'High: stationary' value, i.e. changing from 0.0165 to 0.0202.'β' has a lower γ N than 'Ref.' indicating ITG stabilization due to the increase in β [74].Moreover, the γ N in 'β' is even lower than in 'High: early', implying that the enhanced stabilization due to the increase in β is more than sufficient to compensate for the destabilization of the increase in R/L Ti .
Likewise, γ N in 'β ′ ' relative to γ N in 'Ref.' shows that increasing the normalized local pressure gradient from 0.297 to 0.414 has a slight destabilizing effect.It should be noted that changing the local pressure gradient affects the local equilibrium [75] and the magnetic drift [76].Not shown in figure 18, separate GKW linear runs for these two effects show that the change in local equilibrium destabilizes and the change in magnetic drifts stabilizes.Since the maximum growth rate changes due to each effect are similar in absolute value, 'β ′ ' has slightly increased γ N .
The change in the local parameters for the fast ion species is displayed as 'Fast ions' and shows a relatively weak stabilization of ITG.As shown in section 5.1, the primary change for the fast ion species between 'High: early' and 'High: stationary' comes from the fast ion densities.Therefore, the dilution of the main ions due to the increased fast ion density [77][78][79] is believed to be responsible.Although the stabilization effect is minor in this case, it could be more significant further inside since the fast ion density to the electron density ratio tends to peak on the magnetic axis.While not presented here, additional analysis to check the possible effect of the change in η f ≡ L nf /L Tf [79] confirmed that the η f effect is negligible, where L nf and L Tf are the density and temperature gradient length of the fast ions.
In the case of 'T i /T e ', which represents the change in the ratio of ion and electron temperature, T i /T e , from 0.83 to 1.05, the lowest maximum growth rate is obtained.Although the error bars in the electron temperature are considerable, the stabilizing effect of the increased T i /T e is the most significant compared with the others.
The magnetic shear effect is represented as '(r/q) (dq/dr)'.As shown in figure 2(f ), the magnetic shear at ψ N = 0.3 is increased from 0.38 to 0.45 between 'High: early' and 'High: stationary' by the expanded low magnetic shear region near the magnetic axis.γ N in '(r/q) (dq/dr)' shows the highest maximum growth rate.Additional analysis scanning the magnetic shear from −0.3 to 0.9 showed that γ N continues to increase with the magnetic shear, which is consistent with the previous analysis based on a circular s − α equilibrium model [24].
The linear analysis reveals that the stabilizing effects of ITG from the increased local pressure and T i /T e are more significant than the destabilizing effects from the increased R/L Ti , local pressure gradient, and magnetic shear.Between these two primary stabilizing effects, the enhancement of local pressure is proposed as the trigger for ITG stabilization at 'High: early'.As already seen in section 5.1, the fast ion pressure increases during the performance enhancement due to the decreasing plasma density under CEM activities, contributing to the equilibrium pressure enhancement.Figure 19 shows the increase of the fast ion pressure from 'Low' to 'High: early' and from 'High: early' to 'High: stationary'.Due to the onaxis NBIs, the increase tends to be more significant in the inner core region, which seems to be why the improvement of P th is more pronounced in the inner core region.At the location of analysis, the fast ion pressure accounts for about 30% of the total pressure at 'High: early', which is believed to initiate ITG stabilization.Once the enhanced fast ion pressure initiates ITG stabilization, the weakened R/L Ti stiffness increases the total pressure in the inner core, where the positive feedback loop between ITG stabilization and the total pressure is expected.
On the other hand, there is no reason for T i /T e to increase at 'High: early' and trigger ITG stabilization.No additional heating was applied to change T i /T e in 'High: early' or 'High: stationary'.In addition, the decrease of the plasma density is not favorable for increasing T i /T e in the situation of T i /T e < 1 in 'High: early', where collisional equilibration of thermal energy between ions and electrons is expected to be less with decreasing density.Instead, T i /T e may increase as a result of the ITG stabilization provided by other stabilizing effects, such that T i increases in the inner core region due to increasing R/L Ti .For example, another positive feedback loop for ITG stabilization [80][81][82] can be formed by T i /T e once the ITG mode is stabilized by the local pressure enhanced by the fast ion pressure enhancement.
The stabilization effect of the E × B shear rate is not analyzed since the E × B shear rate in the core region is expected to be similar between 'High: early' and 'High: stationary'.This is because the toroidal rotation gradient, the most dominant component to determine the E × B shear rate in the core region, is similar in the 'High' phases, as shown in figure 2(d).Meanwhile, the similar toroidal rotation profiles are thought to be the reason why the confinement improvement was not achieved in the outer core region.According to the previous work [24], the dominant turbulence stabilization mechanism in the outer core region is the E × B shear rate.Notably, the EM stabilization of ITG by the local fast ion pressure gradient is known to be more significant in nonlinear than in linear analysis, and it becomes more effective at a low magnetic shear regime, such as in hybrid scenarios [10].In this sense, the slightly destabilizing effect of the equilibrium pressure gradient discussed in the above linear analysis could be stabilizing in reality.

Conclusion
This study investigates the origin of the performance improvement observed in KSTAR hybrid scenarios by analyzing several comparative hybrid scenario experiments with different plasma performances.It is found that the performance improvement is mainly due to the edge thermal energy enhanced by the wide and high pedestal formation.In addition, the increase in the fast ion pressure accounts for one third of the total improvement.Although the net change in the core thermal energy is negative due to the reduced core volume resulting from the edge volume expansion, the improved thermal energy confinement in the inner core region of ψ N < 0.4 limits the degradation.
The primary difference between the high-and lowperformance discharges is the presence of vigorous CEM activities obtained at relatively high plasma density and H α .CEM is a coherent edge-localized MHD mode that grows repetitively in pedestal recovery phases.Under CEM activities, particle and heat transport tend to increase in the edge region, leading to pedestal broadening and a continuous density decrease.
Figure 20 summarizes the performance enhancement mechanism proposed in this study.As the pedestal broadens due to the strong CEM activity, the ELM crash becomes delayed compared with the typical ELM cycle without CEMs.Furthermore, as the pedestal height and width continue to increase until the following ELM crash, the maximum achievable edge thermal energy in ELM cycles is enhanced.The ELM crash is weakened after the extended recovery phase.Consequently, the finite thermal energy remains in the pedestal even after the ELM crash, which increases the minimum edge thermal energy in ELM cycles.
Meanwhile, the reduced plasma density under CEM activities extends the beam slowing down time.Under this situation, the number of fast ions in the plasma increases, which consequently increases the total fast ion pressure.Furthermore, the fast ion pressure is primarily enhanced in the inner core region due to the on-axis co-current NBIs in KSTAR, which can initiate the stabilization of ITG, the dominant turbulence in the core region.As the ITG mode is stabilized, the stiffness of R/L Ti is weakened, further increasing the core pressure and forming a positive feedback loop.Similarly, T i /T e is also enhanced by the weakened R/L Ti stiffness, where another positive feedback loop for the ITG stabilization is expected.
This study is meaningful in that it is the first study on the effect of CEM activities on performance improvement in KSTAR hybrid scenario discharges.Experimental observations and simulation results are found to be consistent, which gives us confidence.In terms of advanced scenario development, this study suggests triggering intensive CEM activities as a new controllability of accessing the high-performance hybrid scenario regime.In KSATR, the application of external gas fueling is found to be sufficient to access the CEMdominant high-performance regime.This regime shares similarities with Wide Pedestal QH (WPQH) modes [41,42] in that the edge is enhanced by forming wide and high pedestals.On the other hand, this regime suggests the optimization of ELMs to prevent impurity accumulation, whereas WPQH modes are based on an ELM-free regime.In addition, although the improvement of core confinement by enhancing β and T i /T e is the same as in other advanced core plasmas, it is unique in that the density decrease resulting from CEM activities triggers the improvement.Meanwhile, this regime does not require high toroidal rotation, low density, or a complex control to maintain the regime, which seems relevant for ITER and future machines, such as K-DEMO [83], where sufficiently improved stationary performance without impurity accumulation is required.
On the other hand, there are still issues to be resolved.The most urgent is to find the CEM onset condition and the parameters to intensify its activities.Section 3.2 discusses this subject by focusing on the experimentally observed density effect, but it should be further investigated.For example, there are some cases where one has CEM and the other does not, despite similar plasma densities.Meanwhile, the effect of external gas fueling on CEM activity is partially discussed in section 3.2 and the last part of section 4.However, a key parameter among the possible candidates, such as edge collisionality, pedestal density gradient, and SOL density, needs to be clarified.Fortunately, the edge and SOL density profile measurements with high spatial and temporal resolution are under improvement in KSTAR by utilizing BES and reflectometry, so a more detailed study will be available soon.The improved edge density profile measurements will also allow us to study the particle transport mechanism under CEM activities, which is still unclear.Additionally, we plan to perform the integrated modeling with TRIASSIC [84], where the ELM cycle model with the increased transport due to CEM activities is incorporated into the core transport model.This integrated coreedge simulation will be able to quantitatively investigate the self-consistent performance enhancement process triggered by CEM activities.

Figure 1 .
Figure 1.Overview of shots 25452 (red) and 25458 (blue).I P , P NBI , and P EC are plasma current in 10 2 kA, NBI heating power, and electron cyclotron heating power in MW, respectively.β N is normalized beta.The neutron rate is an uncalibrated value.ne is line-averaged density in 10 19 m −3 , and Hα is H-alpha line intensity with ELM peaks reduced.Vtor,e is edge toroidal rotation in km s −1 .|δB Z (n = 1)| is the amplitude of magnetic fluctuations of n = 1 in G measured by Mirnov coils, where n is the toroidal mode number determined by analyzing an array of fifteen Mirnov coils in the toroidal direction.|δB Z (n = 5 − 8)| is the summed amplitude of magnetic fluctuations of n = 5 − 8. n = 1 and n = 5 − 8 were selected to display amplitude evolutions of the dominant magnetic fluctuations in the core and the edge, respectively.Note that the data after 3.5 s in 25 458 are excluded due to a piggyback experiment conducted from 3.5 s.

Figure 2 .
Figure 2. Electron density (a), electron temperature (b), ion temperature (c), toroidal rotation (d), thermal pressure (solid lines), fast ion pressure (dash-dotted lines), and the gradient of thermal pressure (dashed lines) (e), and safety factor profiles (f ) for three phases, 'Low' (2.4 s in 25458) in blue, 'High: early' (2.4 s in 25452) in red and 'High: stationary' (4.05 s in 25452) in black.(a)-(c) are fitted with an analytic function characterizing the hyperbolic tangent shape for the pedestal structure [43].ψ N is normalized poloidal flux.

Figure 3 .
Figure 3. (a) Core thermal (Core W th in cyan), edge thermal (Edge W th in magenta), and fast ion energy (W f in orange) for the three phases, 'Low' (2.4 s in 285 458), 'High: early' (2.4 s in 25 452) and 'High: stationary' (4.05 s in 25 452).(b) A schematic representation of core and edge thermal energy.W f was calculated by NUBEAM [50].

Figure 4 .
Figure 4. Ion (χ eff i ) and electron (χ eff e ) effective heat diffusivity profiles calculated by power balance analysis for the three phases, 'Low' (2.4 s in 25458), 'High: early' (2.4 s in 25452) and 'High: stationary' (4.05 s in 25452).Note that the error bars were determined by considering the error bars of each measurement.

Figure 5 .
Figure 5. Spectrograms of normalized beam emission intensity fluctuation, δI BES , near the separatrix region (a) and magnetic field fluctuation (b) in the high-performance case, 25 452.(c) Amplitude of magnetic fluctuation of n = 1 (blue), 6 (orange), and 7 (green) in G. δI BES is defined as (I BES − ⟨I BES ⟩) /⟨I BES ⟩, where I BES is beam emission intensity measured by beam emission spectroscopy and <> is time averaging.A toroidal array of Mirnov coils measures the magnetic field fluctuation and its toroidal mode numbers, as introduced in figure 1.The vertical lines in the spectrograms indicate ELM crashes.

Figure 6 .
Figure 6.Ion temperature profile (black) and amplitude of the normalized fluctuation, |δI/⟨I⟩| = |(I − ⟨I⟩) /⟨I⟩|, measured at 'High: early' (2.4 s in 25452), where I is measured local fluctuation and <> means time averaging.I is measured by BES (blue) and ECEI (red), in which only a 15.6 kHz component obtained by applying FFT is displayed to represent CEM.The magenta dashed and yellow dotted lines indicate the last closed flux surface (LCFS) and outer limiter location, respectively.The blue-shaded region represents the BES channels assumed to be susceptible to beam attenuation.

Figure 7 .
Figure 7. (a) Comparing 28 029 (red) and 28 025 (blue).ne is the line-averaged density in 10 19 m −3 , and V GAS in V is the valve voltage to determine the external gas fueling rate.Hα is Hα line intensity.|δI BES | is defined as (I S − ⟨I BES ⟩) /⟨I BES ⟩ filtered for the CEM frequency, where I BES is the beam emission intensity measured by BES and <> is time averaging.j − α diagrams analyzed at 2.320 s (b) and 2.352 s (c), displayed as the vertical dashed lines at t 0 and t 1 in (a), respectively.αmax, jmax, and jsep are the maximum normalized pedestal pressure gradient, the maximum edge current density, and the separatrix current density.The circles in (b) and (c) indicate the equilibria points of each discharge.The solid and the dashed lines in (b) and (c) indicate the stability boundaries defined as γmax/ω * i = 0.25 and γmax/ω A = 0.03, where γmax, ω * i , and ω A are maximum linear growth rate, ion diamagnetic frequency at the center of the pedestal, and the Alfven frequency.V GAS is zero in 28 025.

Figure 8 .
Figure 8. Comparing 26417 (red) and 26416 (blue).ne is line-averaged density in 10 19 m −3 , and V GAS in V is the valve voltage to determine the external gas fueling rate.Hα is Hα line intensity.|δI BES | is defined as (I BES − ⟨I BES ⟩) /⟨I BES ⟩, where I BES is the beam emission intensity measured by BES and <> is time averaging.Wtot is the total stored energy calculated by EFIT in kJ, and Vtor,e is edge toroidal rotation in km s −1 .The external gas fueling rate increases with V GAS .Vtor,e is obtained by the two-Gaussian fitting method[54], which has a lower time resolution than the CEM measurement.

Figure 10 .
Figure 10.CEM fluctuation amplitude measured by BES (|δI BES |), ion temperature (T ion ), and toroidal rotation (Vtor) measured by four channels in the edge region in 28 030, which reveals changes in the edge T ion and Vtor evolution after CEM onset at 2.35 s. 28030 is exactly the same scenario as 28029, shown in section 3.2.

Figure 11 .
Figure 11.Ion temperature, T i , (a), electron density, ne, (b), and pressure, P, and its gradient, dP/dψ N , (c) profiles zoomed near the edge region for 28025 (blue) and 28030 (red).The profiles are picked up at 2.8 s for 28025 and 28030 when the pedestal has been built up under the absence and presence of CEM activities, respectively.P is calculated by multiplying T i and ne.ψ N is normalized poloidal flux.Based on BES and line-averaged density measurements, ne profiles were reconstructed using the model function, assuming a hyperbolic tangent shape for the pedestal.The circles in (a) and (b) are the measured data from CES for T i and the Thomson scattering diagnostics for ne, respectively.

Figure 12 .
Figure 12.Comparison between the measured and the predicted pedestal top (a) and width (b) depending on the presence of CEM activities during the pedestal build-up.The measured parameters shown as Xmeas are from the pedestal ion temperature, while the predicted values as X EPED are from the EPED model.The time slice closest to the following ELM crash was selected for each case in 25452, 25458, 28025, and 28030.The blue squares and the red circles indicate the cases without and with CEM activities, respectively.The dashed lines indicate the y = x line.

Figure 13 .
Figure 13.Electron density (ne) evolution at several ψ N and the summed amplitude of magnetic fluctuation of n = 5 − 8 measured by Mirnov coils (|δB Z (n = 5 − 8)|) in G.The left is the high-performance case, 25452, with CEM, and the right is the low-performance case, 25458, without CEM.Five line-averaged density measurements with different paths were used to reconstruct ne profiles by utilizing the model function and equilibria, as used in figure 2(a).

Figure 14 .
Figure 14.Total stored energy calculated from a diamagnetic loop measurement in kJ (a), the energy increase (in red circles) and the energy decrease (in blue squares) in kJ in each ELM cycle (b), H-alpha line intensity (c), and CEM amplitude measured by Mirnov coil (d) in 26417.(e)-(h) is zoomed-in around 2.9-3.4 s of (a)-(d).

Figure 15 .
Figure 15.Total stored energy (Wtot) in kJ, H-alpha line intensity, magnetic fluctuation amplitude for CEM (|δBz|) in G, pedestal width (∆ width ) in ψ N , pedestal top pressure (P ped ) in kPa, and average pedestal pressure gradient evaluated by the height-to-width ratio (∼∇P ped ) in 28030, the high-performance hybrid scenario discharge introduced in section 3.2.Black squares (A), red circles (B), and blue triangles (C) in the last three rows indicate a typical ELMy H-mode phase, a transiently enhanced performance phase with CEM, and a permanently enhanced performance phase with CEM, respectively.

Figure 17 .
Figure 17.Wavenumber, k θ ρ i , scans for the linear growth rate (γ s −1 , solid lines with squares) and real frequency (ω N , dashed lines with circles) at ψ N = 0.3 in 'High: early' (2.4 s in 25452) and 'High: stationary' (4.05 s in 25452), respectively.k θ and ρ i are wavenumber in the poloidal direction and ion gyro-radius, respectively.The positive value of ω N indicates the ion diamagnetic drift direction.

Figure 18 .
Figure 18.Comparing normalized linear growth rate (γ N ) for the cases with each local parameter varying.All cases are based on 'High: early', shown in red, where the input local parameters are values at ψ N = 0.3 in 'High: early' (2.4 s in 25452).'Ref.' in black is the reference case where the R/L Ti value is substituted by that of 'High: stationary', and the others are from 'High: early' to see the destabilizing effect of enhanced R/L Ti .The others (β, β ′ , Fast ions, T i /Te, and (r/q) (dq/dr)) represent the cases where, based on 'Ref.',another parameter is substituted by that of 'High: stationary' to investigate the effect of each parameter change under the enhanced R/L Ti .'β' in blue indicates the change of normalized pressure, 'β ′ ' in magenta the change of normalized pressure gradient, 'Fast ions' in cyan the change of the parameters of fast ions, 'T i /Te' in orange the change of the ion and electron temperature ratio, and '(r/q) (dq/dr)' in green the change of magnetic shear.

Figure 19 .
Figure 19.The enhancement of the fast ion pressure, ∆P f , between each phase.Changing from 'Low' (2.4 s in 25458) to 'High: early' (2.4 s in 25452) and from 'High: early' to 'High: stationary' (4.05 s in 25452) are shown in red and black, respectively.

Figure 20 .
Figure 20.Mechanism of performance enhancement after the CEM onset.W edge , W f , and Wcore are the edge thermal, the fast ion, and the core thermal energy, respectively.β is the normalized local equilibrium pressure.