Event-by-event analysis of chiral charge separation in $p^{\uparrow}+$Au collisions within an improved AMPT model

We use the string-melting version of the AMPT model [42, 27], with the local-nuclear-scaling refinement of the Lund string-fragmentation parameter $b_{L}$ and the mini-jet cutoff $p_{0}$ introduced in Ref. [44], to simulate $p+$Au collisions at $\sqrt{s_{NN}}=200$ GeV. An event proceeds through four stages: (i) initial conditions generated by HIJING [36, 16], with excited strings melted into partons; (ii) parton transport by the Zhang Parton Cascade (ZPC) [43]; (iii) hadronization through quark coalescence [25, 26]; and (iv) hadronic rescattering described by A Relativistic Transport (ART) model [24].

The proton moves along $+z$ and the Au nucleus along $-z$, each with $E=100$ GeV per nucleon. The proton is transversely polarized along $+x$. Following Ref. [37], four collision-geometry schemes are defined by the orientation of the impact parameter ${\bm{b}}$ relative to the polarization axis. In each event the azimuthal angle $\Psi_{RP}$ is also defined by the oritation of $bb$, which is sampled randomly within a $\pi/2$-wide quadrant centred on the typical direction of the scheme:

• •

  Scheme I: $\Psi_{RP}\in(-\pi/4,\,\pi/4)$ (typical oritation of ${\bm{b}}$ being along $+x$),

• •

  Scheme II: $\Psi_{RP}\in(\pi/4,\,3\pi/4)$ (typical oritation of ${\bm{b}}$ being along $+y$),

• •

  Scheme III: $\Psi_{RP}\in(3\pi/4,\,5\pi/4)$ (typical oritation of ${\bm{b}}$ being along $-x$),

• •

  Scheme IV: $\Psi_{RP}\in(-3\pi/4,\,-\pi/4)$ (typical oritation of ${\bm{b}}$ being along $-y$).

Because the four quadrants together tile the full $2\pi$ exactly once, each event is assigned unambiguously to a single scheme.

In this work, any results within each scheme are averaged over events with random fluctuations of the reaction plane as a real experiment. The magnitude of impact parameter $|{\bm{b}}|=b=1.5$ fm is fixed, in order to concentrate on the gamma-correlator analysis, corresponding approximately to the 0–5% most central bin in $p+$Au.

In our previous work Ref. [39], the parton cross section of $\sigma=10$ mb is used in the stage II of standand AMPT model. This value is appropriate for heavy ion collision system but drives the energy density of small and median collision system to unphysical large magnetude [46]. In the present work we use $\sigma=0.3$ and $0.7$ mb, more in line with the partonic mean free path expected in $p+A$ small system. In the ZPC module, elastic $gg\to gg$ scattering is described by a Debye-screened $t$-channel gluon exchange with differential cross section $d\sigma/dt=9\pi\alpha_{s}^{2}/[2(t-\mu^{2})^{2}]$, which in the high-energy limit $s\gg\mu^{2}$ integrates to [43, 27]

where $\alpha_{s}$ is the strong coupling and $\mu$ is the Debye screening mass. In the standard AMPT parametrisation $\alpha_{s}$ is fixed and the total cross section can be tuned by adjusting $\mu$, so that $\mu^{2}\propto 1/\sigma$. Throughout this work we take $\alpha_{s}=0.33$; the two cross sections then correspond to $\mu=7.612$ fm^-1 ($\sigma=0.3$ mb) and $\mu=4.689$ fm^-1 ($\sigma=0.7$ mb).

Even with these reduced cross sections, the standard AMPT setup still over-produces the Bjorken energy density (see Sec. III), because too many of the initial partons hadronize early and concentrate their transverse energy in a small longitudinal slice at midrapidity. Following the procedure introduced for ¹⁶O$+^{16}$O collisions in Ref. [46], we therefore add an impact-parameter-dependent delay $\Delta\tau(b)$ to the hadron formation time $\tau_{f}$ assigned in the quark-coalescence step. The delay enters as $\tau_{f}\to\tau_{f}+\Delta\tau(b)$ and is applied to every hadron produced from the partonic system; the parton-cascade dynamics in ZPC are left untouched. The net effect is to push the rapid pile-up of transverse energy at $y=0$ later in proper time, so that $\varepsilon_{Bj}(\tau)$ peaks at a lower value, without changing the integrated soft-particle yields.

We parametrize the delay with a sigmoid form,

where

• •

  $A$ (in fm/$c$) is the overall amplitude of the delay; since $B\ll 1$ in our parametrisation, the small-$b$ plateau $\Delta\tau(0)=A/(1+B)$ is essentially equal to $A$, which therefore sets the maximum delay applied at central impact parameters,

• •

  $C$ (in fm) is the characteristic impact-parameter scale over which $\Delta\tau(b)$ falls off; the larger $C$, the more slowly the delay decreases with increasing $b$,

• •

  $B$ (dimensionless) controls the position of the inflection point through $\Delta\tau(b_{*})=A/2$ at $b_{*}=-C\ln B$; smaller $B$ pushes the fall-off region to larger $b$ and thus widens the central plateau.

The three parameters are tuned by hand so that the peak Bjorken energy density at central rapidity stays below $\sim 0.3$ GeV/fm³ in the most central bin while the $b$-dependence of the energy-density fall-off is reproduced; for the two parton cross sections used here we find

• •

  $\sigma=0.3$ mb: $A=1.20$ fm/$c$, $B=0.0143$, $C=0.60$ fm,

• •

  $\sigma=0.7$ mb: $A=1.05$ fm/$c$, $B=0.0143$, $C=0.50$ fm.

In Ref. [39] a fixed 7% momentum exchange was applied uniformly to each event regardless of the local magnetic field. The local $B$ field at the overlap centre, however, fluctuates strongly from event to event and depends on the collision-geometry scheme, so a fixed exchange fraction is at best approximation of the underlying CME dynamics.

We compute the electromagnetic field at the overlap centre event by event from the Lienard–Wiechert potential, using the three-dimensional asymmetric charge profile for the polarized proton and symmetric profiles for the nucleons in the Au nucleus, as in Refs. [45, 37]. The CME-induced charge separation is then injected at the parton level by interchanging the $p_{y}$ components of pairs of co-moving quark and antiquark whose momentum components along ${\bm{B}}$ have opposite signs, mimicking the ${\bm{B}}$-directed chirality-dependent momentum imbalance of the CME: the $p_{y}$ of an upward-moving $\bar{u}$ is swapped with that of a downward-moving $u$, and similarly for the $d/\bar{d}$ pair [29].

The fraction $f$ of such quark pairs that is exchanged in a given event is taken to scale linearly with the magnitude of the magnetic field at the overlap centre,

where $|{\bm{B}}|_{\mathrm{event}}$ is the magnetic-field strength at the overlap centre in the current event and $|{\bm{B}}|_{\mathrm{max}}\simeq 5.6\,m_{\pi}^{2}/e$ is the largest value found across the full sample, taken over all four schemes. The 7% ceiling is fixed by reproducing the $\gamma_{SS}$ correlator in 30–50% Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV [39, 2, 3]. Because the four schemes pick out very different distributions of $|{\bm{B}}|$, Eq. (6) automatically generates four distinct average exchange fractions $\langle f\rangle_{\mathrm{I,II,III,IV}}$, with scheme IV the largest and scheme II the smallest, in line with the geometric prediction $\Delta\gamma_{\mathrm{CME}}\propto B^{2}\cos[2(\Phi_{B}-\Psi_{RP})]$.

The three ingredients introduced above are implemented as local modifications of the public AMPT-SM source code, called as improved AMPT model in this work. The four collision-geometry schemes are realised in hijing1.383_ampt.f: in each event the reaction-plane angle is drawn as $\Psi_{RP}=(\pi/2)\,\xi+\Psi_{0}$, with $\xi$ a uniform random number on $(-1/2,1/2)$ and $\Psi_{0}=0,\pi/2,\pi,3\pi/2$ selecting schemes I–IV. The CME source is implemented inside ZPC (zpc.f): the change subroutine first calls the precomputed Lienard–Wiechert table to obtain $|{\bm{B}}|_{\mathrm{event}}$, evaluates the per-event fraction $f$ of Eq. (6) capped at unity, applies a Knuth shuffle to the lists of co-moving $u$, $\bar{u}$, $d$, $\bar{d}$ quarks, and then exchange the $p_{y}$ of the first $[f\,N_{q}]$ pairs (rounded to the nearest integer) in each flavour list. Finally, the hadron formation-time delay of Eq. (5) is added in linana.f: the parton-to-hadron routine calls a function GET_DELTA_TAU(b) that returns the sigmoid value, which is added to the coalescence formation time of every produced hadron. The two parton cross sections, $\sigma=0.3$ and $0.7$ mb, correspond to the two Debye masses $\mu$ quoted after Eq. (4), and the $(A,B,C)$ parameter sets quoted above are tuned independently for each.

The first validation is the Bjorken energy density as a function of proper time $\tau$,

where $A_{\perp}$ is the transverse overlap area, computed for several impact-parameter bins.

Figures 2 and 3 show $\varepsilon_{Bj}(\tau)$ without the formation-time delay. The most central bin ($b=0$–$1$ fm) reaches peak values of $\sim 2.4$ and $\sim 1.8$ GeV/fm³ for $\sigma=0.3$ and $0.7$ mb respectively, which are unphysically high for $p+$Au. Turning on $\Delta\tau(b)$ with the parameters of Sec. II.2, Figs. 4 and 5, brings the peak central density down to $\sim 0.3$ GeV/fm³, in line with the value expected for small systems [46], while leaving the shape and peripheral behaviour of $\varepsilon_{Bj}(\tau)$ qualitatively unchanged.

As a second, more stringent check we compute $v_{2}(p_{T})$ for charged hadrons in 0–5% $p+$Au using an event-plane procedure designed to mimic the recent PHENIX analysis [5] . The centrality is determined from the charged-hadron multiplicity in the BBC south acceptance ($-3.9<\eta<-3.1$). The event plane is reconstructed from charged hadrons in the FVTX south window ($-3.0<\eta<-1.0$), and its resolution is estimated by the two-sub-event method using $-3.0<\eta<-2.0$ and $-2.0<\eta<-1.0$ as the sub-events. The $v_{2}$ is then measured for charged hadrons at midrapidity, $|\eta|<0.35$.

As shown in Fig. 6, the model with the formation-time delay reproduces the PHENIX data well over $0.4<p_{T}<2.0$ GeV/$c$. The agreement is non-trivial: $\Delta\tau(b)$ was tuned only against the energy density, and the fact that it leaves $v_{2}(p_{T})$ in good agreement with the measurement gives us confidence that the partonic and hadronic phases are reasonably well described by the present setup.

We firstly examine $\gamma_{\alpha\beta}$ as a function of the pair transverse momentum $P_{+}=(p_{T,\alpha}+p_{T,\beta})/2$.

At the initial state (Figs. 7 and 10 corresponding to $\sigma=0.3$ mb and $\sigma=0.7$ mb) all four schemes display a clear splitting between opposite- and same-sign pairs, with a hierarchy of

This ordering reproduces the scheme-by-scheme hierarchy of $B^{2}$ at the overlap centre that we computed in Ref. [37]: in scheme IV the impact parameter points along $-y$, the direction in which the charge asymmetry of the polarized proton constructively enhances $B^{2}$, while scheme II picks up the opposite, destructive geometry. Quantitatively, averaging the $P_{+}$ projection over $0.2\leq P_{+}\leq 1.0$ GeV/$c$, scheme IV yields $|\Delta\gamma|\simeq 5\times 10^{-4}$, scheme I and III about $1.5\times 10^{-4}$, and scheme II about $0.5$–$0.8\times 10^{-4}$, with the same pattern observed in the $\Delta\eta$ projection.

After full evolution (Figs. 9 and 12) the parton cascade and the hadronic rescattering together dissipate most of the initial splitting. The decomposition into the two stages is informative: comparing the after-ZPC parton-level correlator (Figs. 8 and 11) with the initial parton-level one (Figs. 7 and 10), we find that ZPC alone reduces $|\Delta\gamma|$ by only $\sim 10$–$20\%$, whereas the subsequent hadron-coalescence and ART hadronic rescattering bring it down to $\sim 10$–$30\%$ of the initial level. The bulk of the dissipation therefore happens in the hadronic phase rather than during the parton cascade. Despite this large absolute reduction, scheme IV continues to display a clearly visible gap between $\gamma_{OS}$ (positive) and $\gamma_{SS}$ (negative) across the full $P_{+}$ range probed, and the central-value hierarchy $|\Delta\gamma|_{\mathrm{IV}}>|\Delta\gamma|_{\mathrm{I,III}}>|\Delta\gamma|_{\mathrm{II}}$ is preserved at both cross sections, so that the inter-scheme difference $\Delta\gamma_{\mathrm{IV}}-\Delta\gamma_{\mathrm{II}}$ remains a non-trivial observable. The much smaller survival fraction quoted here, compared with the $\sim 40\%$ figure of Ref. [39], originates in the much smaller parton cross sections used in this work ($0.3$ and $0.7$ mb instead of $10$ mb), which shorten the parton-cascade phase and thereby leave a longer effective window for the ART hadronic rescattering to dissipate the charge separation.

The pseudorapidity-gap dependence of $\gamma_{\alpha\beta}$ provides a complementary handle, because the QCD background associated with $v_{2}$ is expected to fall off rapidly with $\Delta\eta=|\eta_{\alpha}-\eta_{\beta}|$ while the CME signal is long-range in $\eta$.

At the initial level (Figs. 13 and 16) the splitting is essentially flat in $\Delta\eta$, which reflects the long-range nature of the source: the $p_{y}$ exchange is applied uniformly along the longitudinal direction, so the resulting correlator carries no intrinsic $\eta$ scale. The same scheme hierarchy as in the $P_{+}$ projection appears here. The after-ZPC stage (Figs. 14 and 17) preserves both the magnitude and the flat profile of the initial state, consistent with the small ZPC-stage suppression observed in the $P_{+}$ projection. After the full evolution (Figs. 15 and 18) the magnitude shrinks substantially, but a clearly visible $\gamma_{OS}/\gamma_{SS}$ gap survives in scheme IV at both $\sigma=0.3$ and $0.7$ mb, with $\gamma_{OS}$ remaining positive and $\gamma_{SS}$ negative across the $\Delta\eta$ range, and a profile that remains broadly flat. In schemes I, II, and III the residual splitting is smaller, but the central-value hierarchy is the same as in the initial state. The persistence of a flat profile under evolution distinguishes $p^{\uparrow}+$Au from Au+Au, where the strong short-range $\Delta\eta$ component of the correlator is dominated by background [29].

Two physics messages emerge. First, the event-by-event implementation of Sec. II.3 automatically inherits the geometric hierarchy of $B^{2}$ across the four schemes, and so produces an initial-state correlator that already encodes the geometric signature of the CME. The same hierarchy persists, although with a much reduced amplitude, through parton cascade and hadronic rescattering. Second, because the background contribution to $\Delta\gamma$ in $p+$Au is small [39], the difference

isolates an essentially pure CME observable. Our results show that this difference remains clearly non-zero after full evolution, with $\Delta\gamma_{\mathrm{IV}}-\Delta\gamma_{\mathrm{II}}$ of order $1$–$2\times 10^{-4}$, which suggests that an experimental measurement with the Zero Degree Calorimeters tagging the impact-parameter orientation should be sensitive to it.

The two cross sections ($\sigma=0.3$ and $0.7$ mb) give qualitatively the same picture, both at the initial level and after evolution. This robustness against a factor-of-two variation in $\sigma$, combined with the validation against the PHENIX $v_{2}(p_{T})$ data, supports the conclusion that the scheme-dependent pattern of $\Delta\gamma$ is dictated by the geometric structure of the magnetic field rather than by the details of the partonic transport.