Decrypting heat transfer

[Yothiel]

So, since I couldn't find any good source on how heat transfer actually worked in this game, my curiosity got the better of me. The decompiled C# DLL might be easy to read and does have some stuff, but most of the tasty bits required me to dive into the disassembly of the C code from the SimDLL. 

In that DLL, I've located three main functions handling most of the following cases:

• Cell/Cell transfers
• Cell/Building transfers
• Cell/Entity transfers (which includes dupes, geysers, mined element chunks and stored items, pretty much everything which is neither a building or a cell)

Most of the following should be valid on both the Oil version and the Automation preview. Reading a proper introduction on thermal transfers may be useful before going over this text, since I'll quickly jump into technical details.

A quick foreword on calculus notations I will use:

• |x| means "absolute value of x". If x > 0, this is equal to x, else it equals -x.
• Δ (delta) has no actual signification, but is used in names  to denote a variation between two steps/elements (for instance, I would be enclined to use "ΔT" if I wanted to put a name on the "T2-T1" value)
• clamp(x, y, z)  assumes that y is lower than z projects the value of x in the interval [y;z]. It is equal to x if x >= y and x <= z, or y if x < y, or z if x > z.

 

Cell/Cell transfers

Okay, so first let's go over the basics. I'll list the fundamental properties and what notation I'll use for them. I'll also indicate the units used in the game, though you should pay them no mind since dimensional correctness is not ensured (especially, most values you'd expect to be watts/joules will actually be used as kilowatts/kilojoules).

Each cell of the game have the following properties:

• a temperature noted T (displayed in K)
• a mass noted m (displayed in kg)
• an insulation factor noted f (dimensionless, varies between 0 (total insulation) and 255 (no insulation). Actually, this value is always 255, excepted insulated tiles which have 2.
• an element (e.g. liquid hydrogen, sandstone, etc)

The same chemical component in two different states (e.g. gas hydrogen and liquid hydrogen) are referred with different elements. Each element has the following properties:

• State information (e.g. gas/liquid/solid of course, but also whether it is unbreakable or temperature-insulated like neutronium)
• a thermal conductivity noted k (the value displayed in W/m/K in the game)
• a specific heat capacity noted c (the value displayed in J/g/K in the game)
• Surface-Area multipliers for interacting with each of the three states gas/liquid/solid. As a rule of thumb, liquids have a multiplier of 25 for interacting with other liquids, and gas also have a multiplier of 25 for interacting with solids. All other multipliers have a value of 1. Let's note them S for now.

If you're being curious, these element values seem to come from some Unity package and are loaded into the simulation, not stored in the DLL.

Now, let's define even MORE values which are derived from the above ones:

• the insulated conductivity noted k' is derived from (gasp) the insulation factor and the thermal conductivity using the following formula: k' = k * (f/255)²
• the heat capacity, noted C is derived from mass and specific heat using the following formula: C = c * m
• the thermal energy, noted Q, derived from the heat capacity and the temperature, using the following formula : Q = C * T

Okay, that should be enough for now.
Let's take a "Cell 1" and a "Cell 2" (I'll suffix each of their properties with their number, e.g. T1 or T2). Here are the steps for determining the heat they transfer during one game tick:

• if any of the elements has the "isolated" flag in its temp (e.g. neutronium), no transfer is done
• if |T2 - T1| < 1, no transfer is done (even when temperature are stabilized, there may still exist a gradient in temperatures)
• A first candidate for the amount of energy transferred during a duration of Δt is given by the formula
  ΔQ = Δt * min(k'1, k'2) * (T2 - T1) * (S1[type of cell 2] * S2[type of cell 1])
  Note that for one game tick, Δt should be 0.25 (4 ticks/second).
• A first clamping is done and results in a new candidate: 
  ΔQ' = sign(T2-T1) * min(|ΔQ|, |Q2-Q1|/8)
• if |ΔQ'| < 1.0E-4, no transfer is done
• Candidates for the new temperatures are computed using the following formula:
  T'1 = T1 + ΔQ' / C1 and T'2 = T2 - ΔQ' / C2 
• if (T2 - T1) * (T'2 - T'1) < 0 (the order of cell temperatures changed), both T'1 and T'2 are affected an equilibrium temperature computed with the formula: T'1 = T'2 = (Q1 + Q2) / (C1 + C2)
• More clamping occurs to prevent temperatures from changing by more than |T2 - T1| / 4:
  if |T'1 - T1| > |T2 - T1| / 4, then T'1 = T1 + (T2 - T1)/4
  likewise if |T2' - T2| > |T2 - T1|, then T'2 = T2 - (T2 - T1)/4
• Final heat transfer is 
  ΔQ'' = sign(ΔQ') * min(|T1' - T1| * C1, |T2' - T2| * C2)
• and final temperatures for the next game tick are:
  T''1 = T1 + ΔQ'' / C1 and T''2 = T2 - ΔQ'' / C2
• and voilà!

So, for some examples with actual figures

1. Cell 1 is 100kg of water at 370K, Cell 2 is 200kg of polluted water at 315K
   ΔQ = 0.25 * min(0.609, 0.58) * (315-370) * 25 * 25 = -4984.375 (note the two "25" surface area multipliers when two liquids interact with each other!)
   With Q1 ~=1.5E5 and Q2 ~=3.6E5, ΔQ' = ΔQ.
   T'1 = 370 -4984.375 / (100 * 4.179) ~= 358K and T'2 = 315 + 4984.375 / (200 * 6) ~= 319.15K
   SInce the temperature changes are within |T2 - T1|/4, no further clamping is needed: these are the final temperatures.
2. Cell 1 is 1kg of hydrogen at 215K, Cell 2 has a metal tile of tungsten. Now, non-permeable tiles will fill the cell with the same material they are composed of, and both the tile and the cell will interact as a single entity, so we have 200kg of tungsten (100kg from the tile and 100kg from the cell) at, let's say, 300K for cell 2.
   ΔQ = 0.25 * min(0.168, 60) * (300 - 215) * 25 * 1 = 89.25
   Again, ΔQ' = ΔQ, so T'1 = 215 + 89.25 / (1 * 2.4) =  252.1875 and T'2 = 300 - 89.25 / (100 * 0.134) = 293.34
   Since |T'1 - T1| > |T2 - T1|/4, we replace it by the value T'1 = 215 + (300 - 215) / 4 = 236.25.
   ΔQ'' = min(|236.25 - 215| * (1 * 2.4), |293.34 - 300| * (100 * 0.134)) = min(51, 89.25) = 51
   Final temperatures are T''1 = 215 + 51/(1 * 2.4) = 236.25 and T''2 = 300 - 51/(200 * 0.134) = 298.097K
3. Cell 1 is abyssalite at 350K and any weight, Cell 2 is granite at 315K and any weight. 
   ΔQ = 0.25 * min(2, 0.00001) * (315 - 350) * 1 * 1 = -8.75E-5
   Since |ΔQ| < 1E-4, no transfer is done.
   Basically, abyssalite will only transfer heat with other solids / liquids until they are within a 40K range of each other. An insulated tile of abyssalite will never transfer heat with another cell (needs a delta of at least 26010K to exchange with a gas, or 650250K for a solid/liquid
4. If you're wondering about that "thermal energy delta" rule, consider the following case:
   Cell 1 is 1kg of hydrogen at 320K, cell 2 is 800kg of gold amalgalm at 6.4K (not a naturally found temperature, but you'll understand the choice)
   Here, Q1 = 2.4 * 1 * 320 = 768, and Q2 = 0.15 * 800 * 6.4 = 768
   This causes ΔQ' to be capped at |Q2 - Q1| / 8 = 0: no transfer is done despite the huge temperature difference and the relatively good conductivities. Making sense to you? Yeah, neither to me

Cell/Building transfers

Okay, now let's add introduce the building!

That term pretty much covers anything you build, but also the ruines you can find prebuilt (even though most of that stuff isn't really doing heat transfer). In the simulation, buildings only interact with the cells they cover, though some specific buildings also interacts with their contents (that's in the C# part of the game and not covered here)

Let's reuse the same properties / notations as the cell, but with a few more definitions:

• left/right/top/bottom coordinates, used for determining the cells interaction with the building. They can be used to define an area, noted A
• a capacity per cell, noted C', and derived from the heat capacity (the usual C = c * m) and area through the formula: C' = C  / 5 / A (note the infamous factor of 5 for all building-related heat transfers)
• An operating heat creation per tick, noted W. This operating heat is displayed in the game as the generated heat by the building, with a number of catches:

1. the displayed value is actually an addition of both the operating heat, and an "excess" heat, the latest being applied for some reason by the C# part of the game and will be listed in the Other transfers part (excess is directly applied to the cell, operating heat is applied to the building which may transfer it to the cell). To see which part is operating and which part is excess and which part is operation, you can hover your cursor over "Heat Production" in the Info / Status tab of an active 
2. the value is displayed as watts, but actually used as kilowatts
3. the value displayed has been multiplied by a 5 factor (strictly speaking this is not the same factor as the one applied to the mass, but the values are probably identical on purpose
   all this considered, you must basically take the value displayed as "normal operation" and divide it by 5 to get the value for W.

Knowing that, there are less rules for cell/building heat transfers. Cell values will be suffixed with 1 (e.g. T1), and building values with 2 (e.g. T2):

• For each game tick and for each building , start by defining an energy accumulator ΔQtot initialized at 0
• For each of the cell inside the left/right/top/bottom area of :
  • If the element id is matching a specific value (Vacuum I'm guessing), no transfer is done, go to the next cell
  • else, let's define the heat capacity Chot. If T1 < T2, Chot = C'2, else Chot = C1
  • Let's also define Tmin = min(T1, T2) and Tmax = max(T1, T2)
  • Candidate value for heat transfer: ΔQ = Δt * k'1 * k2 * Chot * (T1 - T2) * 0.005 * x, where x is a game-global value usually equal to 100 (the only time I saw it with a different value is on the first few ticks after starting/loading a game, in which case it's at a low value like 0.01).
  • candidate new temperatures are clamped to the [Tmin, Tmax] interval: T'1 = clamp(T1 - ΔQ/C1, Tmin, Tmax) and T'2 = clamp(T2 + ΔQ/C'2, Tmin, Tmax) 
  • If (T1 - T2) * (T'1 - T'2) < 0 (order of temperatures changed), values are replaced by the equilibrium temperature T'1 = T'2 = (C1 * T1 + C'2 * T2) / (C1 + C'2)
  • T'1 is set to the cell as its new temperature
  • ΔQtot is increased by C'2 * (T'2 - T2)
• after looping over all the cells for this building, its final temperature for this game tick will be: T''2 = T2 + (ΔQtot + W * Δt)  / (C'2 * A)

A couple examples:

• An active sandstone compost at 275K is sitting on 4 cells, each with 1kg hydrogen at 320k. Its operating heat creation is shown as 5, so W = 1
  The hydrogen is the hot body here, so for each cell, Chot = 1 * 2.4
  ΔQ = 0.25 * 0.168 * 2.9 * 2.4 * (320 - 275) * 0.005 * 100 = 12.6544
  In this case there will be no clamping, and the heat transfer will be the same for each cell, so each cell's temperature will become by T'1 = 320 - 12.6544 / 2.4 = 314.7K (before excess heat is applied), while the compost's temperature will be T''2 = 275 + (4 * 12.6544 + 1 * 0.25) / (0.8 * 800 / 5 / 4 * 4) = 275.4K.
• Take an obsidian pipe at 275K is sitting in 1kg of hydrogen at 320K. Take an obsidian insulated pipe at 275K sitting in another 1kg of hydrogen. 
  Note how there is no place for pipe insulation factor: both pipes will transfer the same heat (though the temperature insulated pipe will have a slower increase because of its increased mass)
• Take the same case as above, but flip the pipe and the hydrogen temperatures. Now, the transferred heat will be proportional to the capacity of the pipes, and the insulated pipe will transfer more heat (though their temperature will be the same)

 

Cell/Entity transfers

Entities have the usual mass / temperature / capacity / conductivity properties, and add the following concepts:

• A Surface Area Multiplier noted S (which, unlike Cell/Cell transfers, does not depend on element state).
• A thickness, noted L

Pretty simple isn't it? Here are the steps. Cell values are suffixed with 1, Entity values with 2.

• An entity exchanges heat with the single cell it is associated to
• If it cannot exchange heat with this cell (i.e. vacuum), it will exchange heat with the cell below, at 1/4 the normal rate
• The candidate heat exchanged for a period Δt is ΔQ = Δt * S2 / L * min(k1, k2) * (T2 - T1) (the 1/4 factor is applied here if the entity exchange heats with the cell below its actual cell)
• if |ΔQ| > |Q1 - Q2| / 1000, ΔQ = sign(ΔQ) * |Q1-Q2| / 1000 (yeah, the nonsensical clamping based on thermal energy difference is back!)
• The candidate new temperatures are T'1 = T1 + ΔQ / C1 and T'2 = T2 - ΔQ / C2
• if (T1 - T2) * (T'1 - T'2) < 0, the equilibrium temperature is back: T'1 = T'2 = (Q1 + Q2) / (C1 + C2)
• That's over!

Now, there is an exception: items stored in a fridge (even unpowered) or a compost will interact with a virtual heat source of 277.15K (fridge) or 348.15K (compost), and will do so with a set conductivity of 1000. The cell will not receive / lose heat from that exchange (technically, you could destroy heat by repetitively putting muckroot in a fridge and getting it out, but that would not be very efficient)

I think most entities have a S value of 1, and a thickness of 0.001. Some exceptions I've spotted:

• wheezewort have a S value of 10
• Duplicants have a thickness depending on their clothes and the displayed value after conversion to meters (for 0.2cm, L = 0.002). There is one catch however that I managed to spot while debugging the code: thickness from the default clothes is actually not counted (so basically, even the cool vest will keep duplicants warmer than default clothes)

 

Other transfers

While this does seem a lot, there still are many cases not accounted yet when it comes to heat. Here is a non-exhaustive list:

• pipe/content transfer: uses a standard formula for determining the heat transfer, which is pretty much equivalent to the Cell/Entity one without the thermal energy clamping if you take S / L = 50 (source: C# code)
• building/cell transfer of "excess" heat: each cell receives the excess heat, divided by the area of the building, and this transferred heat is proportionally scaled down if the cell as less than 1.5kg of content (source: C# code)
• Building-specific heat creations/transfers (e.g. aquatuners, refineries, etc). Each case is different, and once again can be checked in human-readable decompiled C#, so I won't bother listing them.
• Heat exchange accompanying movement/exchange of liquids/gas: I can't tell you a thing about that, though it's probably the part which was responsible for the infamous "dripping bug"
• Heat propagation in gas: when a gas cell receives heat, it will send some of it to the cell above if it contains the same type of gas. If the cell is losing heat, it will instead propagate some of that loss to the cell below if it contains the same type of gas. This propagation continues until the next cell contains a different element, or if the transferred heat is too low. Source: experimentation in debug mode.
• any heat creation / destruction linked to element creation / destruction / conversion

 

Okay, I think that covers pretty much I've been able to siphon so far. While most of this information may only be marginally useful, it may help understand some oddities such as:

• the "insulated" part of insulated pipes
• the excellent results of very conductive tiles (and the steel previously created by closed doors!) for exchanging heat with the far less conductive granite pipes.
• the actual effect of insulated tile (woosh this stuff is almost as good as abyssalite!)
• etc

Feel free to add any information & let me know if something is unclear or if an error found its way in this huge text (which wouldn't surprise me tbh), and have fun trying (and probably failing) to size any radiator / cooling system!

[eloy2030]

39 minutes ago, Yothiel said:

If you're wondering about that "thermal energy delta" rule, consider the following case:
Cell 1 is 1kg of hydrogen at 320K, cell 2 is 800kg of gold amalgalm at 6.4K (not a naturally found temperature, but you'll understand the choice)
Here, Q1 = 2.4 * 1 * 320 = 768, and Q2 = 0.15 * 800 * 6.4 = 768
This causes ΔQ' to be capped at |Q2 - Q1| / 8 = 0: no transfer is done despite the huge temperature difference and the relatively good conductivities. Making sense to you? Yeah, neither to me

Oh! So, THAT´S why!

nice post, man!

Edit: I was wondering, though, why does it transfer with more tiles involved.

I hope I´m not making a stupid question...

[GrindThisGame]

You should write science textbooks

[ithilelda]

6 hours ago, Yothiel said:

Now, let's define even MORE values which are derived from the above ones:

• the insulated conductivity noted k' is derived from (gasp) the insulation factor and the thermal conductivity using the following formula: k' = k * (f/255)²

Well then, I guess the 400kg cost of insulated tiles is worthy now. Previously I thought it is 20% the conductivity, but from the formula it is less than 0.1 thousandth. I don't have to search for abyssalites just to get my room insulated!

 

6 hours ago, Yothiel said:

a capacity per cell, noted C', and derived from the mass, specific heat and area through the formula: C' = C * m / 5 / A (note the infamous factor of 5 for all building-related heat transfers)

I think you made a typo here. Capital C is heat capacity and has mass already multiplied in it. The formula should be C' = C / 5 / A.

 

6 hours ago, Yothiel said:

Candidate value for heat transfer: ΔQ = Δt * k'1 * k2 * Chot * (T1 - T2) * 0.005 * x, where x is a game-global value usually equal to 100 (the only time I saw it with a different value is on the first few ticks after starting/loading a game, in which case it's at a low value like 0.01).

My brain failed to comprehend where in the entire universe will the right side of the equation result in a unit of Joules... How did they came up with this?

 

7 hours ago, Yothiel said:

the excellent results of very conductive tiles (and the steel previously created by closed doors!) for exchanging heat with the far less conductive granite pipes.

Is this because the above formula has k of both materials in it (rather than picking the minimum as in cell-cell interactions)?

 

7 hours ago, Yothiel said:

Note how there is no place for pipe insulation factor: both pipes will transfer the same heat (though the temperature insulated pipe will have a slower increase because of its increased mass)

Shouldn't insulated pipe also have a k' that's 2/255 square times of k in the cell-building transfer? Or does that 2/255 factor only apply to tiles? The 20% insulation is actually the pipe insulation factor? If that's the case, then the 400kg insulated gas pipe is absolutely useless. It doesn't provide that much of a difference, and costs 375 kg more material to build...

Last question, are thermshift plates 3x3, or do they only cover the 4 orthogonal tiles?

Anyway, great work! This is serious information!

[Yothiel]

28 minutes ago, ithilelda said:

Well then, I guess the 400kg cost of insulated tiles is worthy now. Previously I thought it is 20% the conductivity, but from the formula it is less than 0.1 thousandth. I don't have to search for abyssalites just to get my room insulated!

Yup, even with granite, an insulated tile will "only" have an actual conductivity 20x higher than a standard abyssalite tile

Quote

I think you made a typo here. Capital C is heat capacity and has mass already multiplied in it. The formula should be C' = C / 5 / A.

Fixed, thanks!

Quote

My brain failed to comprehend where in the entire universe will the right side of the equation result in a unit of Joules... How did they came up with this?

 

they probably played with formulae until they reached something which did not feel too broken ingame.
Actually that's the very formula which sprung me into action, as I was dumbfounded by some tests in debug mode where buildings with a higher conductivity would exchange less heat than the same building with lower conductivity in the same conditions (because capacity also played a role, but only if the building was hotter than its cell).

Quote

Is this because the above formula has k of both materials in it (rather than picking the minimum as in cell-cell interactions)?

Exact. And that would get quite high if you built 

Quote

Shouldn't insulated pipe also have a k' that's 2/255 square times of k in the cell-building transfer? Or does that 2/255 factor only apply to tiles? The 20% insulation is actually the pipe insulation factor? If that's the case, then the 400kg insulated gas pipe is absolutely useless. It doesn't provide that much of a difference, and costs 375 kg more material to build...

That's the thing, there is no k' computation for buildings. Actually I don't think the pipe insulation defined in C# files reaches the simulation at all, only the insulated tile which implements the "Insulator" template passes it.

Quote

Last question, are thermshift plates 3x3, or do they only cover the 4 orthogonal tiles?

3x3 area. As the code is right now, only rectangular areas are supported for building heat exchange.

[Risu]

45 minutes ago, Yothiel said:

That's the thing, there is no k' computation for buildings. Actually I don't think the pipe insulation defined in C# files reaches the simulation at all, only the insulated tile which implements the "Insulator" template passes it.

Yeah the insulation is only used to change the color in the overlay. As you might have found, the Insulator class sends the cell index of the tile, not a reference to the building. Suspect adding the component to the insulated pipe would have some amusing results.
 

[Yothiel]

3 hours ago, Risu said:

Yeah the insulation is only used to change the color in the overlay. As you might have found, the Insulator class sends the cell index of the tile, not a reference to the building. Suspect adding the component to the insulated pipe would have some amusing results.
 

Exactly what I thought! You could nerf the insulation of an insulated tile by constructing an insulated pipe into it, then remove the remaining insulation by deconstructing the pipe. Or you could make a pretty decent heat barrier just by building a circuit of insulated pipes...

[Cilya]

Very nice work with this reverse engineering Yothiel.

I'm completely frightened and probably disgusted a bit by what you have found... No matter why we couldn't understand heat transfer. It makes no sense. The only way it can makes sense is your interpretation : the formula have been tweaked with very dirty hacks to try to remove unwanted behaviors in the game. But it really gives the bad impression that the modifications are clueless. The Chot and the distance between thermal energies just doesn't have any sense. Some tweaks have been done for some heat transfers not some others. I thought the core of the game engine was robust, now I'm kind of depressed...

[Mjello]

In the attempt not to insult anyone I will ... can it recompile?

There is something I want to change

[AnonymousRetard]

2 hours ago, Mjello said:

In the attempt not to insult anyone I will ... can it recompile?

There is something I want to change

It might be possible if you try hard enough. But I think you might get into legal problems if you actually try to spread modifed versions or modified parts of the game that way. I don't think the game is supposed to be open source so it's probably not a good idea to publically post too much of the games code either if you have decompiled it.

But if you have the correct knowledge and don't talk about it publically you will be able to do whatever you want of course... 

I can't speak for the developers but if I was them I would be ok with all the information Yothiel provided in the original post, but I don't think they would like information being spread about how to decompile/recompile any parts of the game.

[Mjello]

1 minute ago, AnonymousRetard said:

It might be possible if you try hard enough. But I think you might get into legal problems if you actually try to spread modifed versions or modified parts of the game that way. I don't think the game is supposed to be open source so it's probably not a good idea to publically post too much of the games code either if you have decompiled it.

But if you have the correct knowledge and don't talk about it publically you will be able to do whatever you want of course... 

I can't speak for the developers but if I was them I would be ok with all the information Yothiel provided in the original post, but I don't think they would like information being spread about how to decompile/recompile any parts of the game.

Yeah that makes completely good sense.

It is proprietary and their only way to protect their work.

Well I can allready say there is no easy way to recompile. It is well protected.

[Octyabr]

7 hours ago, Mjello said:

In the attempt not to insult anyone I will ... can it recompile?

There is something I want to change

If you are using ILSpy to peek into the code, you could use reflexil to directly edit IL code.
Now that I see in its page, it "also supports C#/VB.NET code injection." Is that new or did I just never noticed?

[ithilelda]

Thanks for the detailed response! I guess the cell-building formula is what it is and I'll just have to admit that it works in ONI universe. A valuable lesson is that I don't have to waste resource on insulated pipes now. Certainly insulated tiles with igneous rock/obsidian and abbysalite pipes to go!

[Yothiel]

On 16/11/2017 at 2:50 PM, Cilya said:

Very nice work with this reverse engineering Yothiel.

I'm completely frightened and probably disgusted a bit by what you have found... No matter why we couldn't understand heat transfer. It makes no sense. The only way it can makes sense is your interpretation : the formula have been tweaked with very dirty hacks to try to remove unwanted behaviors in the game. But it really gives the bad impression that the modifications are clueless. The Chot and the distance between thermal energies just doesn't have any sense. Some tweaks have been done for some heat transfers not some others. I thought the core of the game engine was robust, now I'm kind of depressed...

Now now, don't get me feeling bad about sharing those results ^^

To be honest, as a dev myself, it doesn't really surprise me. In some complex parts of a program, some strange unwanted behaviors can lurk for years before being discovered by chance. Or they may have been voluntarily introduced and left as is until you discover some new cases being broken by them. 

In that situation, I don't think the thermal energy delta causes much issues since it creates an instable equilibrium point which is pretty much impossible to reach normally, and it may even help with some cases I haven't tought about. As for the building / cell transfer, it doesn't really break the simulation either, even though the lack of a real insulation factor for buildings is quite misleading (but this is not really related to the rest of the formula with Chot & cie)

 

On 16/11/2017 at 4:32 PM, Mjello said:

In the attempt not to insult anyone I will ... can it recompile?

There is something I want to change

Technically, everything is possible on code running on your machine. But in reality, you will definitely not be able to do something simple about it.

The main reason is that there is no "real" automatic decompilation for C/C++ code. It may be possible to decompile to some source code that would recompile with the same behavior, but it would be unreadable and a single non-trivial change in that "decompiled" source would break the result. You really need human insight to understand what the program actually does, and the process isn't exactly quick either.

Bottom line, don't expect an easy way to tamper with the simulation code until the game dev decide to offer some way to do it.

[avc15]

part of a model like this is thinking how computationally and design intensive it'd be for them to use the real formulas from physics, and a bit unnecessary. They have to make some simplifications to get a game that will play at any kind of speed. OK, so some things will not be perfect in a simplified model.

So now you've gone and explained heat conduction. Has anyone deciphered the mechanics of convection yet? In other words, how does the game decide how each "packet" of gas winds up in its particular cell, when two low pressure packets combine into one, what motivates them to move about?

Same thing could be looked at for both liquids and gases.

Do warmer fluids rise/colder fluids fall?

Do materials have latent heat? (example, does the process of evaporation absorb heat, or just changing temps)

[Cilya]

But it is not simple ! There are a lot of clamping which introduce corner cases.

18 hours ago, avc15 said:

part of a model like this is thinking how computationally and design intensive it'd be for them to use the real formulas from physics, and a bit unnecessary.

They are using real formula. They did a discrete simulation of physics which is completely standard and perfectly fine. None of that is a problem. You are missing the point.

[UristMcKerman]

Honestly I'm surprised that it's a C# code and not some sort of shader language. Works quite fast given the size of the map.

[Cilya]

The map is not that big. I play on a 512x512 map, with 1/4 of a million. The simulation updates 4 times per seconds, so it's only 1 million cell updates per second. For a modern computer, it's not much.

[Risu]

1 hour ago, UristMcKerman said:

Honestly I'm surprised that it's a C# code and not some sort of shader language. Works quite fast given the size of the map.

The bulk of all the computations are done in a native DLL. Plain C code.
 

[avc15]

14 hours ago, Cilya said:

They are using real formula.

Hmm, not really. There are some significant simplifications in this model.

Hence the need for clamping - the results probably needed some serious sanity added to them in testing. I'm willing to bet that the clamping and enveloping were necessary to eliminate all kinds of odd behaviors.

I don't think I'm missing your point at all.

 

--

Regardless, I'm still more interested in the other things. Is there latent heat? How does convection work? Do hot fluids flow upwards relative to cold fluids?

[Cilya]

The heat capacity and heat conductions are real formula. I understand that some clamping have been added, and I understand why. For instance, the absence of transfer when the difference in temperature is small enough can improve the performances of the simulation. But this rounding can be added to the real formula as an error, and we can estimate how far from the real formula we are. These are mathematically valid approximations. But for the clamping around the difference of thermal energy differences, I can't make any sense of it. If you have insights about it, I'm really interested.

[avc15]

6 hours ago, Cilya said:

The heat capacity and heat conductions are real formula.

No, really, it's not. It mimics reality somewhat at best.

Simplification #1 (of several)

They're treating all heat exchanges as if they were taking place between identical materials (completely neglecting the junction and the 2nd material's heat transfer coefficient).

--

the extra steps some people were railing about are probably to keep undesired behaviors out. Undesired behaviors will come from the peculiar and simplified nature of their model. And the contrived geometry of things, the contrived materials properties, probably also a desire to slow things down adequately for a real player to manage. 

[Cilya]

1 hour ago, avc15 said:

It mimics reality somewhat at best.

I'm not saying anything else. There is no doubt they are not taking every possible parameter. Neither is there a doubt that the formula they used  come from models which are valid at some point. My criticism comes from the fact that, despite doing valid approximations, they also do approximations which doesn't seem to make any sense. Again, if you have any insight about the approximations I've pointed out, I'll be more than happy to change my mind.

[Pumbo]

They should be more transparent how the temperature works in game, because the temperature control is the only challange on this game.