betaFeedback

WorksPhysics and Astronomy

Moment tracking and their coordinate transformations for macroparticles with an application to plasmas around black holes

Abstract

Particle-in-cell codes usually represent large groups of particles as a single macroparticle. These codes are computationally efficient but lose information about the internal structure of the macroparticle. To improve the accuracy of these codes, this work presents a method in which, as well as tracking the macroparticle, the moments of the macroparticle are also tracked. Although the equations needed to track these moments are known, the coordinate transformations for moments where the space and time coordinates are mixed cannot be calculated using the standard method for representing moments. These coordinate transformations are important in astrophysical plasma, where there is no preferred coordinate system. This work uses the language of Schwartz distributions to calculate the coordinate transformations of moments. Both the moment tracking and coordinate transformation equations are tested by modelling the motion of uncharged particles in a circular orbit around a black hole in both Schwarzschild and Kruskal-Szekeres coordinates. Numerical testing shows that the error in tracking moments is small, and scales quadratically. This error can be improved by including higher order moments. By choosing an appropriate method for using these moments to deposit the charge back onto the grid, a full particle-in-cell code can be developed.

Introduction

In numerical simulations involving the dynamics of a large number of particles, for example, in a plasma, it is impossible to track the dynamics of every ion and electron. In particle-in-cell (PIC) codes, the plasma is modelled using macroparticles, where each simulated macroparticle represents a large number of actual particles. To improve the accuracy of a PIC code, there are two options: use more macroparticles, or give each macroparticle more information. A macroparticle typically has only a position and a velocity. This paper presents a different method, in which a macroparticle represents the moments of a group of particles [1, 2, 3, 4, 5] (Figure 1). Such a model may be more efficient in cases where a large number of particles can be accurately modelled by only a small number of macroparticles and their moments, and the electromagnetic field does not vary much across the extent of the macroparticle. Additionally, in cases where the dominating interactive force can be calculated from the Liénard-Wiechert potential, the electromagnetic fields can be calculated directly from the moments of the macroparticle [6].

Figure 1. Tracking several individual particles compared to a macroparticle with moments. The 5 particles (the black lines) in the left diagram are replaced by a single macroparticle (the orange line) in the right diagram. By tracking the moments, quantities such as the difference between the centre of charge of the 5 particles and the position of the macroparticle (the first order moment, represented by the horizontal blue arrows), and the variance in position of the particles (the second order moment, represented by the green error bars) can be tracked.

There are several existing methods for tracking moments: the code MERLIN implements a transition matrix approach for particle accelerators [4] and a continuous model has been developed through a Hamiltonian approach [5]. Moment tracking can also be done continuously by differentiating the definition of a moment and using the Vlasov equation [3, 7]. In plasma physics, the concept of moment tracking is often used to describe hybrid-Vlasov approaches, where individual species of the plasma may be modelled through their bulk properties [8, 9, 10]. Moments in the context of hybrid-Vlasov approaches are constructed by integrating over velocity space only and can be interpreted as physical quantities such as temperature and pressure. In this work, moments shall be constructed by integrating over both velocity and position space, giving related, but different quantities. The transport of moments is also used in wider fields where the Liouville equation holds, such as particle nucleation [11], crystal growth [12], and nuclear collisions [13].

To perfectly track the moments, an infinite set of moments is required. This is because, as is shown in this paper, higher order moments generate lower order moments. Tracking more moments is more computationally expensive, as these moments come with additional computational work (both more differential equations to solve each time step and more memory usage per macroparticle). This increased computational work is balanced by reducing the total number of macroparticles in the simulation. At the quadrupole (second) order there are 33 equations to solve each time step (21 quadrupole moments, 6 dipole (first order) moments, 3 components of velocity and 3 components of position). This expands to 215 equations if the expansion is carried out to the hexadecapole level (fourth order) (see Table I). To make moment tracking computationally feasible, a truncation is required. This truncation is the highest order of moments considered, above which the contribution from higher order moments are neglected. Care must be taken to ensure the truncation is of as low an order as possible to minimise computational load, whilst also ensuring that neglecting the higher order moments does not significantly impact accuracy. This article calculates all quantities to quadrupole order, although all results presented can be generalised to arbitrary order.

Order of moments trackedInformation knownNumber of differential equations to solveNumber of differential equations to solve relative to a standard macro-particle
Monopole (zeroth order)Xμ,UμX^{\mu}, U^{\mu}61
Dipole (first order)Xμ,Uμ,VaX^{\mu}, U^{\mu}, V^{a}122
Quadrupole (second order)Xμ,Uμ,Va,VabX^{\mu}, U^{\mu}, V^{a}, V^{ab}335.5
Octopole (third order)Xμ,Uμ,Va,Vab,VabcX^{\mu}, U^{\mu}, V^{a}, V^{ab}, V^{abc}8914.83
Hexadecapole (fourth order)Xμ,Uμ,Va,Vab,Vabc,VabcdX^{\mu}, U^{\mu}, V^{a}, V^{ab}, V^{abc}, V^{abcd}21535.83

Table I. The amount of information in a macroparticle that also tracks moments compared to a standard macroparticle. This increased information results in more memory usage and a larger number of first order differential equations to solve. Since there are 6 differential equations per time step for a standard macroparticle, the fourth column is the third column divided by six.

Recently there has been a focus on using PIC codes to model the dynamics of plasma around black holes [14, 15, 16, 17]. Such plasmas may be important in active galactic nuclei, pulsars and gamma-ray bursts [17, 18, 19]. Ref. [20] contains a full review of these studies. In such systems, there is not a preferred choice of coordinate system i.e. when modelling a static uncharged black hole, there is a choice to work in Schwarzschild coordinates, or Kruskal-Szekeres coordinates, amongst others. Because there is a choice in coordinate systems, it is useful to be able to transform between different coordinate systems, especially where the time and space coordinates are mixed together (such as the coordinate transform between Kruskal-Szekeres and Schwarzschild coordinates, shown in Figure 2). Coordinate transformations that mix space and time coordinates also appear in particle accelerators. When simulating the motion of a linearly accelerating bunch in a particle accelerator, the transformation into the instantaneous rest frame of an accelerating bunch mixes space and time coordinates (this transformation is similar to the one shown in Figure 2), so the spacetime coordinate transformations presented in this work are necessary.

Figure 2. A spacetime diagram in Kruskal-Szekeres coordinates of a particle travelling at constant rr in Schwarzschild coordinates (the orange hyperbola). The diagonal blue lines are time slicings in Schwarzschild coordinates (constant tt), and the horizontal purple lines are time slicings in Kruskal-Szekeres coordinates (constant TT). The dashed black line is the event horizon. The moments in a given coordinate system have no components in the direction of the time slicing in that coordinate system.

The moments of a macroparticle depend on the choice of time slicing (Figure 3). In all coordinate systems considered in this article, the global coordinate time will be used as

Figure 3. An example of two different time slicings. When taking moments, the time slicing is the hypersurface the moments are integrated over. By using a different time slicing (e.g. either the horizontal dashed purple lines tt or the curved dashed blue lines t^\hat{t}), the distance between the centre of the macroparticle (the orange line) and the nearby particles the macroparticle is representing (the green contours) is different, so different moments will be found.

the time slicing. This time slicing is a foliation given by spatial hypersurfaces of constant coordinate time. In general, different coordinate systems will give different time slicings. This means when transforming between coordinate systems that mix temporal and spatial coordinates, for example, transforming between Schwarzschild and Kruskal-Szekeres coordinates, the time slicing will change (Figure 2). This time slicing may be the global Killing timelike vector in relativistic scenarios, or the lab time in particle accelerators. Another possible time slicing is the backward light cone of an observer, which is the frame used when making astrophysical observations. A choice of time slicing commonly used in modelling plasma around black holes is found in the fiducial observer (FIDO) scheme, where the time slicing is given relative to local moving observers [21, 22]. A fourth possible time slicing is to take all the vectors orthogonal to the velocity of the world line. By using geodesics to propagate these to the world line, these can be used for the Dixon representation of a multipole [23].

One of the key results of this article is the formula for the coordinate transformation of moments between coordinate systems that mix space and time coordinates. The standard integral representation of moments cannot transform moments between coordinate systems where the time slicing changes. Simply transforming the moments into the new coordinate system does not take into account the change in time slicing. If one were to use the standard representation of moments, an “effective” coordinate transformation can be performed. This is the process of reconstructing the distribution function from the moments, transforming the distribution function into the new coordinate system and retaking the moments in the new coordinate system. In order to find the coordinate transformations that change the time slicing, a different representation of moments must be used, in terms of Schwartz distributions [24, 25]. In this representation of moments, the moments are transformed into the new coordinate system, then projected onto the new time slicing to find the full coordinate transformation. In a standard particle-in-cell code there are three stages: the particle pusher, charge and current deposition, and the field updater. This work will focus only on the particle pusher stage of the PIC code, showing how tracking the moments of the group of particles represented by the macroparticle gives extra structure to the information within the macroparticle. There will be no consideration of particle-particle interactions. Particles will only be affected by external fields. This approach means it is not meaningful to talk about the cells at this stage. Future work will consider how to deposit the charge and current associated with the extra structure of the moments onto the grid. This future work will examine the appropriate cell size needed for this deposition.

This article calculates the moment tracking equations and coordinate transformations for arbitrary coordinate systems. The structure of this article is as follows: Section II introduces the Vlasov equation, which describes the dynamics of collisionless charged particles. Section III introduces moments, and the existing methods for modelling them. Section IV introduces a distributional representation of moments in terms of derivatives of Dirac delta functions and uses this representation to show the time evolution of moments. Section V finds the coordinate transformations for the moments, which is the main result of this work. Section VI shows a computational model of uncharged particles circling a black hole to validate the theory at a proof-of-concept level. Section VII uses the language of differential geometry and de Rham currents to present a geometric interpretation of the moment differential equations and finds the coordinate transformations of the moments through this language. Section VIII discusses the considerations that will need to be made in the development of a method that uses the moments to deposit charge and current onto the grid, in order to develop a full PIC code. We conclude in section IX by discussing other interesting direction of future investigation. The appendix contains the more technical proofs needed to establish the theory.

Notation of indices

To study the dynamics of plasma, calculations are performed in 7 dimensional phase space. Because of this, several different summation conventions for indices are needed: summations over just space, just velocity, space and velocity, time and space but not velocity, and space, time and velocity. In order to account for all these different summations, this article uses the convention that Latin indices a,b,ca,b,c represent summations over 0,…,60,\ldots,6; Greek indices μ,ν,ρ\mu,\nu,\rho represent summations from 0,1,2,30,1,2,3; and underlined index means there is no summation over the 00 index, i.e. a‾=1,…,6\underline{a}=1,\ldots,6 and μ‾=1,2,3\underline{\mu}=1,2,3. The coordinates used are (t,xμ,uμ)(t,x^\mu,u^\mu), where tt is the global time, xμx^\mu is a space coordinate, and uμu^\mu is a spatial component of the 4-velocity. The notation (ξa)(\xi^a) will be used to represent a general coordinate, such that

ξ0=t,ξμ=xμ,ξμ+3=uμ(1)\xi^0=t,\qquad\xi^\mu=x^\mu,\qquad\xi^{\mu+3}=u^\mu \tag*{(1)}

i.e. ξ4=u1\xi^4=u^1. Dimensions are used such that c=G=1c=G=1. The notation ∫p\int_p will be used to represent the evaluation of a function at a point, i.e. ∫pf\int_p f is evaluated at pp. After their introduction, the arguments of functions will not be written, unless needed for emphasis.

The Vlasov Equation

7 Dimensional phase space

To calculated the complex dynamics of plasmas, calculations are performed in 7 dimensional phase space. These dimensions are time, 3 spatial dimensions and 3 proper velocity (4-velocity) dimensions, where tt is the global coordinate time. As previously stated, these will be represented in coordinates as (t,xμ,uμ)(t,x^\mu,u^\mu). The time component of 4-velocity, u0(t,xμ,uμ)u^0(t,x^\mu,u^\mu), is a function in phase space, defined as the positive solution to the quadratic equation

gμνuμuν=−1(2)g_{\mu\nu}u^\mu u^\nu=-1 \tag*{(2)}

where g(t,xμ)g(t,x^\mu) is the metric with signature (−,+,+,+)(-,+,+,+). In the specific example of Minkowski spacetime with Cartesian coordinates, u0u^0 is the Lorentz factor in special relativity, given by

u0=(1+∑μ=13uμuμ)12.(3)u^0=\left(1+\sum_{\mu=1}^{3}u^\mu u^\mu\right)^{\frac{1}{2}}. \tag*{(3)}

The Vlasov equation

Consider the dynamics of a group of charged particles, each with charge qq and mass mm with distribution function f(t,xμ,uμ)f(t,x^\mu,u^\mu). By assuming these particles are collisionless, they can be modelled by the Vlasov equation. To find the Vlasov equation, consider an arbitrary spacetime with metric gμν(t,xμ)g_{\mu\nu}(t,x^\mu), Christoffel symbols Γνρμ(t,xμ)\Gamma^\mu_{\nu\rho}(t,x^\mu) and electromagnetic 2-form Fμν(t,xμ)F_{\mu\nu}(t,x^\mu). Let C(t)C(t) be a world line parameterised by tt, the same parameter as the global time ξ0\xi^0. Let η(t)\eta(t) be the prolongation of C(t)C(t), that is, a curve such that

η0(t)=C0(t)=t,ημ(t)=Cμ(t),ημ+3(t)=dCμ(t)dt.(4)\eta^0(t)=C^0(t)=t,\qquad\eta^\mu(t)=C^\mu(t),\qquad\eta^{\mu+3}(t)=\frac{dC^\mu(t)}{dt}. \tag*{(4)}

For a given value of tt, η(t)\eta(t) is a point in 7-dimensional phase space. In terms of coordinate functions (t,xμ,uμ)(t,x^\mu,u^\mu), this can be represented as

η0(t)=C0(t)=t,ημ(t)=Cμ(t)=xμ∣η(t),ημ+3=dCμ(t)dt=uμu0∣η(t).(5)\eta^0(t)=C^0(t)=t,\qquad\eta^\mu(t)=C^\mu(t)=x^\mu|_{\eta(t)},\qquad\eta^{\mu+3}=\frac{dC^\mu(t)}{dt}=\frac{u^\mu}{u^0}\bigg|_{\eta(t)}. \tag*{(5)}

In general, η(t)\eta(t) is a curve in position-velocity phase space. In this work η(t)\eta(t) will be the curve moments are taken around. In particle accelerators, the ideal orbit is a natural choice for η(t)\eta(t). This choice does not exist in plasmas; in these cases choices for η(t)\eta(t) could be the position of the macroparticle, or a trajectory based on the initial centre of charge of the macroparticle.

The Vlasov vector field, WW, is the integral curves of η\eta, defined as

Wa∣η(t)=ddtηa(t).(6)W^a|_{\eta(t)}=\frac{d}{dt}\eta^a(t). \tag*{(6)}

Finding these derivatives gives 7 ODEs,

W0∣η=dC0(t)dt∣η=1,Wμ∣η=dCμ(t)dt∣η=uμu0∣η(7)W^0|_\eta=\frac{dC^0(t)}{dt}\bigg|_\eta=1,\qquad W^\mu|_\eta=\frac{dC^\mu(t)}{dt}\bigg|_\eta=\frac{u^\mu}{u^0}\bigg|_\eta \tag*{(7)}
Wμ+3∣η=d2Cμ(t)dt2.W^{\mu+3}|_\eta=\frac{d^2C^\mu(t)}{dt^2}.

The acceleration can be found using the pregeodesic equation combined with the (pre-)Lorentz force,

d2Cμ(t)dt2∣η+(ΓνρμdCν(t)dtdCρ(t)dt)∣η=(κdCμ(t)dt)∣η+(qm1u0FνρgμνdCρ(t)dt)∣η(8)\left.\frac{d^2C^\mu(t)}{dt^2}\right|_\eta+\left.\left(\Gamma^\mu_{\nu\rho}\frac{dC^\nu(t)}{dt}\frac{dC^\rho(t)}{dt}\right)\right|_\eta =\left.\left(\kappa\frac{dC^\mu(t)}{dt}\right)\right|_\eta+\left.\left(\frac{q}{m}\frac{1}{u^0}F_{\nu\rho}g^{\mu\nu}\frac{dC^\rho(t)}{dt}\right)\right|_\eta \tag*{(8)}

where the 1/u01/u^0 term in the Lorentz force arises when coordinate time time is the parameterisation, rather than proper time, and κ(t)\kappa(t) arises from the parameterisation. κ(t)\kappa(t) can be found by noting C0=tC^0=t, hence d2C0/dt2=0d^2C^0/dt^2=0, solving this gives

κ=Γνρ0dCν(t)dtdCρ(t)dt−qm1u0Fνρg0νdCρ(t)dt.(9)\kappa=\Gamma^0_{\nu\rho}\frac{dC^\nu(t)}{dt}\frac{dC^\rho(t)}{dt}-\frac{q}{m}\frac{1}{u^0}F_{\nu\rho}g^{0\nu}\frac{dC^\rho(t)}{dt}. \tag*{(9)}

From this, solving equation (7) for all η\eta gives

References

  1. [1]A. Dymnikov and E. Perelshtein, Nucl. Instrum. Methods 148, 567 (1983).
  2. [2]P. J. Channell, IEEE Trans. Nucl. Sci. 30, 2607 (1983).
  3. [3]P. J. Channell, L. M. Healy, and W. P. Lysenko, IEEE Trans. Nucl. Sci. 32, 2565 (1985).
  4. [4]R. B. Appleby, R. J. Barlow, D. Krücker, J. Molson, S. Rowan, S. Tygier, H. Rafique, N. Walker, and A. Wolski, Comput. Phys. Commun. 271, 108204 (2022).
  5. [5]B. A. Shadwick and J. S. Wurtele, CERN Doc. Serv. (1999).
  6. [6]J. R. Ellis, J. Math. Phys. 7, 1185 (1966).
  7. [7]W. Ackermann and T. Weiland, in Proc. ICAP 2006 (Chamonix, France, 2006).
  8. [8]M. Palmroth, U. Ganse, Y. Pfau-Kempf, M. Battarbee, L. Turc, T. Brito, M. Grandin, S. Hoilijoki, A. Sandroos, and S. von Alfthan, Living Rev. Comput. Astrophys. 4, 1 (2018).
  9. [9]R. J. Mason, J. Comput. Phys. 41, 233 (1981).
  10. [10]A. R. Bell, J. R. Davies, S. Guerin, and H. Ruhl, Plasma Phys. Control. Fusion 39, 653 (1997).
  11. [11]H. M. Hulburt and S. Katz, Chem. Eng. Sci. 19, 555 (1964).
  12. [12]H. Grubin, T. Govindan, J. Kreskovsky, and M. Stroscio, Solid-State Electron. 36, 1697 (1993).
  13. [13]M. A. Stroscio, Superlattices Microstruct. 2, 83 (1986).
  14. [14]B. Cerutti, G. R. Werner, D. A. Uzdensky, and M. C. Begelman, ApJ 770, 147 (2013).
  15. [15]B. Crinquand, Ph.D. thesis, Université Grenoble Alpes [2020-....] (2021).
  16. [16]B. Crinquand, B. Cerutti, A. Philippov, K. Parfrey, and G. Dubus, Phys. Rev. Lett. 124, 145101 (2020).
  17. [17]K. Parfrey, A. Philippov, and B. Cerutti, Phys. Rev. Lett. 122, 035101 (2019).
  18. [18]J. Rosenberg and F. Ebrahimi, Astrophys. J. Lett. 920, L29 (2021).
  19. [19]A. A. Philippov and A. Spitkovsky, Astrophys. J. 855, 94 (2018).
  20. [20]K. Nishikawa, I. Duţan, C. Köhn, and Y. Mizuno, Living Rev. Comput. Astrophys. 7, 1 (2021).
  21. [21]K. S. Thorne and D. MacDonald, Mon. Not. R. Astron. Soc. 198, 339 (1982).
  22. [22]I. Y. Dodin and N. J. Fisch, Phys. Plasmas 17, 112118 (2010).
  23. [23]J. Gratus and S. Talaganis, Class. Quantum Grav. 40, 085012 (2023).
  24. [24]J. Gratus and T. Banaszek, Proc. R. Soc. Math. Phys. Eng. Sci. 474, 20170652 (2018).
  25. [25]J. Gratus, P. Pinto, and S. Talaganis, Class. Quantum Grav. 38, 035011 (2020).
  26. [26]S. L. Ginzburg, V. F. Dyachenko, Yu. N. Orlov, N. N. Fimin, and V. M. Chetkekin, Comput. Math. and Math. Phys. 56, 1611 (2016).
  27. [27]An integral over the unit hyperboloid must contain the measure det(g)/u0. There is a choice of making either f a density of weight 1, or to put the measure in the integrands. This article uses the convention that f is a scalar density of weight 1.
  28. [28]J. R. Ellis, Math. Proc. Camb. Philos. Soc. 78, 145 (1975).
  29. [29]A. Warwick and J. Gratus (2023). "Dataset of moment coordinate transformations", Zenodo. http://dx.doi.org/10.5281/zenodo.8082181.
  30. [30]V. John, I. Angelov, A. A. Öncül, and D. Thèvenin, Chem. Eng. Sci. 62, 2890 (2007).
  31. [31]H. Cramér, Princeton Mathematical Series No. 9 (university press, Princeton, 1946).
  32. [32]J. Villasenor and O. Buneman, Comput. Phys. Commun. 69, 306 (1992).
  33. [33]T. Zh. Esirkepov, Comput. Phys. Commun. 135, 144 (2001).
  34. [34]C. K. Birdsall and A. B. Langdon, Plasma Physics Via Computer Simulation (McGraw-Hill, 1985).
  35. [35]R. Hockney, J. Comput. Phys. 8, 19 (1971).
  36. [36]J. U. Brackbill, J. Comput. Phys. 317, 405 (2016).
  37. [37]D. C. Barnes and L. Chacón, Comput. Phys. Commun. 258, 107560 (2021).
  38. [38]A. B. Langdon, J. Comput. Phys. 6, 247 (1970).
  39. [39]C. P. Ridgers, C. S. Brady, R. Duclous, J. G. Kirk, K. Bennett, T. D. Arber, and A. R. Bell, Phys. Plasmas 20, 056701 (2013).
  40. [40]D. Wu, X. T. He, W. Yu, and S. Fritsche, High Power Laser Sci. Eng. 6, e50 (2018/ed).
  41. [41]B. Shadwick, J. Wurtele, A. Sessler, C. Celata, and P. Lee, in Proc. 1999 Part. Accel. Conf. Cat No99CH36366, Vol. 3 (1999) pp. 1716–1718 vol.3.

Paper details

vlnc:W_eejrdh2r
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents