Thomas-Fermi impact ionization model #1754

n01r · 2017-01-09T15:47:36Z

This pull request adds the Thomas-Fermi impact/collisional ionization model to PIConGPU.

The model is implemented using the field ionization framework introducing grid-mapped and particle-position-interpolated density and "temperature" fields as input parameters. For more information see this issue.

The implementation also makes use of multiple FieldTmp slots (#1703).

Preliminary compile- and runtime tests have been successful and show good agreement of HDF5 output and analytical expectation of charge states.

I propose to do a follow-up PR with a new ThinFoil example to free up the LaserWakefield example.

ToDo

Clean up the commit history and regroup into meaningful units

~~Please refrain from commenting on small inconsistencies until the above box has been checked.~~

Replace round2Nearest(avgChargeState) by RNG generated distribution between floor and ceil (Update 2017-01-10)

n01r · 2017-01-09T15:49:39Z

@ax3l Does it make sense to add this PR to the Collisional Ionization Methods project?

ax3l · 2017-01-09T15:54:06Z

yes sure, done!

I propose to do a follow-up PR with a new ThinFoil example to free up the LaserWakefield example.

yes, sounds good!

n01r · 2017-01-10T11:52:04Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                /** @TODO replace the static_cast<float_64> by casts to float_X
+                 * or leave out entirely and compute everything in PIConGPU scaled units
+                 */
+                float_64 const densityUnit = static_cast<float_64>(particleToGrid::derivedAttributes::Density().getUnit()[0]);


Do calculations in numerically stable scaled units if possible.

Seems not really possible in general right now. Maybe at a later point.

n01r · 2017-01-10T11:54:46Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                float_64  x = TFAlpha * math::pow(Q,TFBeta_temp);
+
+                /* Thomas-Fermi average ionization state */
+                float_X ZStar = static_cast<float_X>(protonNumber * x / (float_64(1.) + x + math::sqrt(float_64(1.) + float_64(2.)*x)));


Let's see if we can go from float_64 to float_X.

@ax3l could you suggest a way to calculate this in more numerically stable units?
I don't really see it right now. Normally this is fine because you need to arrive at the correct units in the end and can just multiply after all calculations back into SI. But since this is a fitted formula what do I do here?

E.g. with a terms like math::pow(R,B) and math::pow(R,C)

yep, it's nasty but possible.
first start by checking the dimensions of the constants/fit parameters (and add them). We can then normalize those quantities in formula accordingly. If the formula is SI already, then it's just simple scaling of the constants before doing the calc.

after offline discussion: So do we agree on keeping the calculation in double precision for now since there is no easy way of retracing how and with which units this was fitted?

n01r · 2017-01-10T12:58:49Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                /** determine the new number of bound electrons from the TF ionization state by rounding to nearest charge state
+                 * @TODO introduce partial macroparticle ionization / ionization distribution at some point
+                 */
+                float_X newBoundElectrons = protonNumber - float_X(math::float2int_rn(ZStar));


Follow this up with a commit that calculates

/* works only for positive charge states */ float_X chargeState = math::floor(ZStar) + static_cast<float_X>(randNr < math::modf(ZStar)) float_X newBoundElectrons = protonNumber - chargeState;

where randNr is a random number.
If for example ZStar = 3.6 this would yield a charge state of 4+ in 60% and a charge state of 3+ in 40% of the cases.

n01r · 2017-01-10T13:07:34Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                /** convert kinetic energy in J to "temperature" in eV by assuming an ideal electron gas
+                 * E_kin = 3/2 k*T
+                 */
+                float_64 temperature = kinEnergy * UNITCONV_Joule_to_keV * float_64(1.e3) * float_64(2./3.);


What do we actually want to call an "electron temperature"?

@ax3l , @HighIander : ideas?

ToDo: Implement cut-off at 50 keV (?) / Make the cut-off a user-definable parameter.

Well, the cut-off is an important problem and there is a very recent scientific discussion about exactly this topic here: Phys. Rev. Lett. 113, 255001 (2014) and in the corresponding comment and reply.
Summarizing, I think a 30 keV cutoff as default is OK for now, but one should allow for a user-defined value and (later) even better would allow for a real fit with a Maxwell to the low-energy part or getting the thermal velocity by any other means "on-the-fly".

Yes, we should keep the assumption that one can derive a "temperature" in the simple model from the mean with user-defined cut-off and should then rather check against one of the many extensions of TF, if one is better suited non-thermal distributions for our use cases (see the references from last week in David Salzmann, 1998, Ch. 2.4).

n01r · 2017-01-24T10:30:24Z

Go ahead @ax3l , @psychocoderHPC you can review from here. 👍 .
No need to introduce new feature parts all in this PR.

ax3l · 2017-01-27T17:57:31Z

@n01r compile error, e.g.

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp(136):
  error: no instance of function template "PMacc::algorithms::math::fmod" matches the argument list
    argument types are: (picongpu::float_X)

psychocoderHPC · 2017-01-27T18:07:01Z

`fmod` takes two arguments. Am 27. Januar 2017 18:57:31 MEZ, schrieb Axel Huebl <notifications@github.com>:

…

@n01r compile error, e.g. ``` src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp(136): error: no instance of function template "PMacc::algorithms::math::fmod" matches the argument list argument types are: (picongpu::float_X) ``` -- You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub: #1754 (comment)

-- Diese Nachricht wurde von meinem Android-Mobiltelefon mit K-9 Mail gesendet.

n01r · 2017-02-03T12:37:04Z

Yeah, corrected that offline already and had it in a git stash which I then forgot 🤦‍♂️

n01r · 2017-02-06T13:41:57Z

@ax3l checks are passing, could you review this week, please?

ax3l · 2017-02-07T19:11:33Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+/** \file AlgorithmThomasFermi.hpp
+ *
+ * IONIZATION ALGORITHM for the Thomas-Fermi model
+ *


can you add the original paper and maybe other materials you found relevant for your implementation at this point?

-> done so in the .def file, very good!

please remove the empty line and keep one block. had problems with adding newlines in the \file description in the last days.

Write like this please:

/** \file AlgorithmThomasFermi.hpp * * IONIZATION ALGORITHM for the Thomas-Fermi model * - implements the calculation of average ionization degree and changes charge states * by decreasing the number of bound electrons * - is called with the IONIZATION MODEL, specifically by setting the flag in @see speciesDefinition.param */

ax3l

Implementation and test results look good to me, it would be nice if you could upload a test result and describe what you did offline to the PR, too.

Sadly it's a bit hard to find out the correct scaling without crawling into details of the fit that was used and badly documented in the original paper. But this will be ok for now.

A few inline comments on wording, modern style for new code and a few questions. Please check your include orders, I already changed a few in #1800 so you can take a look (described there) and adjust here in the new files accordingly.

Thanks for the clean implementation and testing! ✨

ax3l · 2017-02-07T19:13:10Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+     * \brief calculation for the Thomas-Fermi pressure ionization model
+     *
+     * This model uses local density and "temperature" values as input
+     * parameters. To be able to speak of a "temperature" an equilibrium state


Phrasing suggestion: ~~To be able to speak of~~ -> A physical temperature requires a defined equilibrium state ...

ax3l · 2017-02-07T19:15:00Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+     * would be required. Typical high power laser-plasma interaction is highly
+     * non-equilibrated, though. The name "temperature" is kept to illustrate
+     * the origination from the Thomas-Fermi model. It is nevertheless
+     * more accurate to think of it as an averaged kinetic energy.


... to think of it as an averaged kinetic energy, which is not backed by the Model and should therefore be only used with a certain suspicion in such Non-LTE scenarios.

ax3l · 2017-02-07T19:16:37Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+        {
+
+            /* @TODO replace the float_64 with float_X and make sure the values are scaled to PIConGPU units */
+            const float_64 protonNumber = GetAtomicNumbers<ParticleType>::type::numberOfProtons;


I think both lines can be constexpr

ax3l · 2017-02-07T19:16:52Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+            if (chargeState < protonNumber)
+            {
+                /* atomic mass number (usually A) A = N + Z */
+                float_64 massNumber = neutronNumber + protonNumber;


should also be a constexpr

ax3l · 2017-02-07T19:17:45Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                float_64 const densityUnit = static_cast<float_64>(particleToGrid::derivedAttributes::Density().getUnit()[0]);
+                float_64 const kinEnergyDensityUnit = static_cast<float_64>(particleToGrid::derivedAttributes::EnergyDensity().getUnit()[0]);
+                /* convert from kinetic energy density to average kinetic energy per particle */
+                float_64 kinEnergy = (kinEnergyDensity / density) * (kinEnergyDensityUnit / densityUnit);


const float_64 kinEnergy

ax3l · 2017-02-07T19:28:19Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/ThomasFermi_Impl.hpp

+
+#pragma once
+
+#include <boost/type_traits/integral_constant.hpp>


please move this to the bottom, just before stdlib includes

there are no stdlib includes here, right? ^^; so this will be placed at the last place in the include list

ax3l · 2017-02-07T19:29:31Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/ThomasFermi_Impl.hpp

+                /* alias for the single macro-particle */
+                auto particle = ionFrame[localIdx];
+                /* particle position, used for field-to-particle interpolation */
+                floatD_X pos = particle[position_];


ax3l · 2017-02-07T19:32:00Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/ThomasFermi_Impl.hpp

+            typedef typename RNGFactory::GetRandomType<Distribution>::type RandomGen;
+            RandomGen randomGen;
+
+            typedef MappingDesc::SuperCellSize TVec;


that is a very nasty renaming, because you use it below as actual mapping to get a 3D index inside the supercell.
not sure if the rename is needed at all, otherwise just make it shorter, e.g., using SuperCellSize = MappingDesc::SuperCellSize;

ax3l · 2017-02-07T19:32:34Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/ThomasFermi_Impl.hpp

+    struct ThomasFermi_Impl
+    {
+
+        typedef T_DestSpecies DestSpecies;


if you don't mind, please use the C++11 using NewType = OtherType; since it's more readable for new code.

ax3l · 2017-02-07T19:34:27Z

src/picongpu/include/particles/ionization/byCollision/collisionalIonizationCalc.def

+namespace ionization
+{
+
+    struct AlgorithmNone;


do you really need that?

That is the fallback like in fieldIonizationCalc.def. There should be other ways to do a fallback, though.

yep, both should be removed. if no algo is used, just don't add the flag to the Particle class in the species. can be done in a follow-up PR for the field ioniz.

n01r · 2017-02-13T14:11:42Z

Merge conflicts in ionizerConfig.param - also still recent fix to ionizationMethods.hpp in #1824.
Need to rebase this against dev before finally uploading some test results and the eventual merge.

n01r · 2017-02-14T13:59:40Z

I force-pushed the rebased and amended commit history. The branch was compile- and runtested for a couple of simulation steps. Before I will test the more complicated setup of a spatial temperature gradient I created a little steady-state scenario.

It is in an unphysical regime but that doesn't matter for the model functionality.

We have:

neutral nitrogen gas at a homogenous density of 1e25 m^-3
electrons at the same density with an initial temperature of 17eV
no pushers
no current
ThomasFermi ionization model (of course)
3 time steps (already 1 is enough)

We expect an average charge state of roughly 4.5.

That leaves us with an expected average number of boundElectrons equal to 2.5.
This is the statistics overview of the boundElectrons attribute of the 4,194,291 nitrogen ions after the first time step. ⬇️

Looks successful to me.

ax3l

Slight changes and rebase pls :)

ax3l · 2017-02-17T14:58:40Z

examples/LaserWakefield/include/simulation_defines/param/speciesDefinition.param

@@ -121,6 +122,10 @@ struct Hydrogen
 *                -> circularly polarized lasers
 * - Keldysh : Keldysh ionization model
 *
+ * - ThomasFermi : statistical impact ionization based on Thomas-Fermi
+ *                 atomic model


after rebase, please add this section to ionizerConfig.param now

ax3l · 2017-02-17T15:02:10Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                float_64 constexpr nAvogadro = SI::N_AVOGADRO;
+                float_64 constexpr convM3ToCM3 = 1.e6;
+
+                float_64 const convToMassDensity = densityUnit * massNumber / nAvogadro / convM3ToCM3;


massNumber / nAvogadro / convM3ToCM3 can also be expressed as constexpr :-)

constexpr float_64 cleverNameYouChoose = massNumber / nAvogadro / convM3ToCM3; float_64 const convToMassDensity = densityUnit * cleverNameYouChoose;

ax3l · 2017-02-17T15:03:27Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                float_64 const convToMassDensity = densityUnit * massNumber / nAvogadro / convM3ToCM3;
+                float_64 const massDensity = density * convToMassDensity;
+
+                float_64 constexpr atomicTimesMassNumber = protonNumber * massNumber;


review comment not applied fully. Please write

float_64 constexpr invAtomicTimesMassNumber = float_64(1.) / (protonNumber * massNumber); float_64 const R = massDensity * invAtomicTimesMassNumber;

instead.

Sorry, you're right. Done.

ax3l · 2017-02-17T15:05:01Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+                float_64 const C = TFC1 * T_F + TFC2;
+
+                /* requires mass density in g/cm^3 */
+                float_64 constexpr nAvogadro = SI::N_AVOGADRO;


please keep constexpr as the first argument. constexpr is not a type specifier which can be read right to left but is a specifier for the expression you are writing currently. discussed again today with @psychocoderHPC

example difference to const: not all arguments on the right hand side need to be const to define a const.
but: to define a constexpr variable, the expression itself needs to be a constant expression.

Thanks for the clarification!
Done!

ax3l · 2017-02-17T15:06:47Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/ThomasFermi.def

@@ -0,0 +1,77 @@
+/**


/** -> /*

ax3l · 2017-02-17T15:07:35Z

src/picongpu/include/particles/ionization/byCollision/ionizers.hpp

@@ -0,0 +1,30 @@
+/**


/** -> /*

ax3l · 2017-02-17T15:10:42Z

src/picongpu/include/simulation_defines/param/speciesDefinition.param

@@ -138,6 +139,10 @@ struct Hydrogen
 *                -> circularly polarized lasers
 * - Keldysh : Keldysh ionization model
 *
+ * - ThomasFermi : statistical impact ionization based on Thomas-Fermi


after rebase, please add this block in ionizerConfig.param

ax3l · 2017-02-17T15:12:09Z

src/picongpu/include/particles/ionization/byCollision/ionizers.hpp

+
+#pragma once
+
+/** \file ionizers.hpp


above #pragma once pls

ax3l · 2017-02-17T15:13:48Z

src/picongpu/include/particles/ionization/byCollision/ThomasFermi/AlgorithmThomasFermi.hpp

+/** \file AlgorithmThomasFermi.hpp
+ *
+ * IONIZATION ALGORITHM for the Thomas-Fermi model
+ *


please remove the empty line and keep one block. had problems with adding newlines in the \file description in the last days.

Write like this please:

/** \file AlgorithmThomasFermi.hpp * * IONIZATION ALGORITHM for the Thomas-Fermi model * - implements the calculation of average ionization degree and changes charge states * by decreasing the number of bound electrons * - is called with the IONIZATION MODEL, specifically by setting the flag in @see speciesDefinition.param */

ax3l · 2017-02-17T15:14:07Z

src/picongpu/include/particles/ionization/byCollision/ionizers.def

+
+#pragma once
+
+/* \file ionizers.def


please above #pragma once

done, also changed to /** \file ...

ax3l · 2017-02-17T15:16:44Z

src/picongpu/include/simulation_defines/param/ionizerConfig.param

+{
+namespace ionization
+{
+


plase add a thomasFermi namespace, after rebase you see that you can skip the particles namespace here

I added a particles namespace as well - but just for the TF fitting parameters for now to not break the methods which use the atomic and effective atomic numbers.

I still had to add a lot of thomasFermi:: though. (Unless every model gets its own namespace.)

ax3l · 2017-02-17T15:17:48Z

src/picongpu/include/simulation_defines/param/ionizerConfig.param

+     * \url http://www.sciencedirect.com/science/article/pii/S0065219908601451
+     * doi:10.1016/S0065-2199(08)60145-1
+     */
+    float_X constexpr TFAlpha = 14.3139;


keep constexpr left pls (see above)

Adds a statistical collisional/impact ionization model based on the Thomas-Fermi atomic model. Files: - AlgorithmThomasFermi.hpp: calculates an average charge state - TFFittingParameters.def: provides the numerical fitting parameters - ThomasFermi.def: defines the model for use in speciesDefinition.param - ThomasFermi_Impl.hpp: prepares necessary fields, caches data, calls the calculation

Adds forward declarations and hints to implementation to prepare the model for being chosen in the params.

Depending on the effective charge state `ZStar` which is usually a floating point value the new charge state for the particle will be determined by `floor(ZStar)` plus the result of a Monte-Carlo process. (0 or 1)

n01r · 2017-02-22T10:28:11Z

offline talk: refactor all (new) typedefs to using

- Replaced `/** Copyright` with `/* Copyright` in license headers to fix doxygen - moved all `constexpr` to the very left again - put the TF fitting parameters into `namespace thomasFermi` - replaced all `typedef`s by C++11 `using`

n01r · 2017-02-22T13:49:33Z

offline talk: ~~dc.releaseData~~ the RNG used by ThomasFermi_Impl.hpp still produces a seg fault after the simulation has finished but this will be changed later in DataConnector.hpp
update #1886

Apart from that the runtime checks seem okay after rebase and changes.

n01r · 2017-02-22T15:21:05Z

Merrrrrr-ge me .... harrrrrr-der!

ax3l

Wonderful, well done!

little follow-up to ComputationalRadiationPhysics#1754 adding a missing include that caused compile issues

ax3l self-requested a review January 9, 2017 15:51

ax3l added component: core in PIConGPU (core application) feature labels Jan 9, 2017

ax3l self-assigned this Jan 9, 2017

ax3l requested a review from psychocoderHPC January 9, 2017 15:51

ax3l added this to the Next Stable: 0.3.0 / 1.0.0 milestone Jan 9, 2017

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch 3 times, most recently from ef5e08a to ed83deb Compare January 10, 2017 10:58

n01r commented Jan 10, 2017

View reviewed changes

n01r mentioned this pull request Jan 10, 2017

Implementing Thomas-Fermi pressure ionization #1444

Closed

2 tasks

n01r commented Jan 10, 2017

View reviewed changes

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch from ed83deb to cb6c60e Compare January 24, 2017 10:14

ax3l changed the title ~~[WIP] Add Thomas-Fermi impact ionization model~~ Add Thomas-Fermi impact ionization model Jan 24, 2017

ax3l changed the title ~~Add Thomas-Fermi impact ionization model~~ Thomas-Fermi impact ionization model Jan 24, 2017

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch from cb6c60e to c2d52bb Compare February 3, 2017 13:01

ax3l reviewed Feb 7, 2017

View reviewed changes

ax3l requested changes Feb 7, 2017

View reviewed changes

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch from 7933c3b to 7b31142 Compare February 13, 2017 14:09

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch from 7b31142 to 35099ff Compare February 13, 2017 14:37

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch 2 times, most recently from 1e4043b to 89691fa Compare February 14, 2017 13:32

ax3l requested changes Feb 17, 2017

View reviewed changes

ax3l reviewed Feb 17, 2017

View reviewed changes

Marco Garten added 6 commits February 21, 2017 17:56

Add forward declarations for Thomas-Fermi

758bb81

Adds forward declarations and hints to implementation to prepare the model for being chosen in the params.

Make ThomasFermi eligible in the parameterSets

a195979

Add random number generation to Thomas-Fermi

bcc830c

Depending on the effective charge state `ZStar` which is usually a floating point value the new charge state for the particle will be determined by `floor(ZStar)` plus the result of a Monte-Carlo process. (0 or 1)

Implement Reviewer suggestions

e3ea835

Move file documentation for AlgorithmThomasFermi

3be23e2

Implement reviewer suggestions Vol. 2

bf1999e

- Replaced `/** Copyright` with `/* Copyright` in license headers to fix doxygen - moved all `constexpr` to the very left again - put the TF fitting parameters into `namespace thomasFermi` - replaced all `typedef`s by C++11 `using`

n01r force-pushed the topic-ThomasFermiNewFieldTmp branch from 89691fa to bf1999e Compare February 22, 2017 12:59

ax3l approved these changes Feb 22, 2017

View reviewed changes

ax3l removed the request for review from psychocoderHPC February 22, 2017 15:26

ax3l merged commit c9fa051 into ComputationalRadiationPhysics:dev Feb 22, 2017

n01r deleted the topic-ThomasFermiNewFieldTmp branch February 22, 2017 15:28

ax3l added a commit to ax3l/picongpu that referenced this pull request Apr 27, 2017

ThomasFermi: fix missing include

3f7cfba

little follow-up to ComputationalRadiationPhysics#1754 adding a missing include that caused compile issues

ax3l mentioned this pull request Apr 27, 2017

ThomasFermi: fix missing include #2003

Merged

ax3l added a commit to ax3l/picongpu that referenced this pull request Apr 28, 2017

ThomasFermi: fix missing include

187e5e7

little follow-up to ComputationalRadiationPhysics#1754 adding a missing include that caused compile issues

ax3l added a commit to ax3l/picongpu that referenced this pull request May 11, 2017

ThomasFermi: fix missing include

544ca07

little follow-up to ComputationalRadiationPhysics#1754 adding a missing include that caused compile issues


		#pragma once

		#include <boost/type_traits/integral_constant.hpp>

Thomas-Fermi impact ionization model #1754

Thomas-Fermi impact ionization model #1754

Conversation

n01r commented Jan 9, 2017 • edited Loading

This pull request adds the Thomas-Fermi impact/collisional ionization model to PIConGPU.

ToDo

n01r commented Jan 9, 2017

ax3l commented Jan 9, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

n01r Jan 27, 2017 • edited Loading

Choose a reason for hiding this comment

n01r Jan 10, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

n01r commented Jan 24, 2017

ax3l commented Jan 27, 2017

psychocoderHPC commented Jan 27, 2017 via email

n01r commented Feb 3, 2017

n01r commented Feb 6, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ax3l left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ax3l Feb 13, 2017 • edited Loading

Choose a reason for hiding this comment

n01r commented Feb 13, 2017

n01r commented Feb 14, 2017

ax3l left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ax3l Feb 17, 2017 • edited by n01r Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

n01r Feb 22, 2017 • edited Loading

Choose a reason for hiding this comment

ax3l Feb 17, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

n01r commented Feb 22, 2017 • edited by ax3l Loading

n01r commented Feb 22, 2017 • edited Loading

n01r commented Feb 22, 2017

ax3l left a comment

Choose a reason for hiding this comment

n01r commented Jan 9, 2017 •

edited

Loading

n01r Jan 27, 2017 •

edited

Loading

n01r Jan 10, 2017 •

edited

Loading

ax3l Feb 13, 2017 •

edited

Loading

ax3l Feb 17, 2017 •

edited by n01r

Loading

n01r Feb 22, 2017 •

edited

Loading

ax3l Feb 17, 2017 •

edited

Loading

n01r commented Feb 22, 2017 •

edited by ax3l

Loading

n01r commented Feb 22, 2017 •

edited

Loading