In other languages: Deutsch Français 日本語 Polski Русский Українська 简体中文

Uranium processing: Difference between revisions

From Official Factorio Wiki
Jump to navigation Jump to search
Tag: Reverted
(Precise mathematical analysis moved to User:Fluffyliberta, see User_talk:Fluffyliberta. I suggest to post it on the forums or on reddit and link that in the see also.)
Tag: Manual revert
 
(9 intermediate revisions by one other user not shown)
Line 36: Line 36:
|-
|-
|}
|}
== Precise Mathematical Analysis ==
To determine the exact number of [[centrifuge]]s required for maintaining [[reactor]] stability without using the [[Kovarex enrichment process]], we need to employ probability theory. This analysis will help us estimate the probability of obtaining the required amount of [[uranium-235]] and the number of [[centrifuge]]s needed.
The primary goal of these calculations is to derive a formula that can be easily applied and sustainable for predictions despite advancements in [[technologies]], game version updates, or in-game developments and modifications.
=== Variables ===
Let's introduce some variables that we'll need in further calculations
* ''<code>T</code> - overall time of a [[reactor]] working (in seconds*)''
* ''<code>t</code> - time of a [[reactor]] cycle (in seconds*)''
* ''<code>c</code> - consumption of [[uranium-235]] per reactor cycle''
* ''<code>k</code> - [[centrifuge]]s number (our target value for calculations)''
* ''<code>p</code> - probability of [[uranium-235]] production per centrifuge cycle''
* ''<code>n</code> - overall number of [[centrifuge]] cycles''
* ''<code>&tau;</code> - [[centrifuge]] time per cycle''
* ''<code>&alpha;</code> - desired probability of having success number [[uranium-235]] to sustain all <code>T</code> time of [[reactor]] stable work**''
<nowiki>*</nowiki> Measured in seconds or in-game [[tick]]s, this does not affect the general reasoning
<nowiki>**</nowiki> Known as the [[https://en.wikipedia.org/wiki/Confidence_interval confidence level]]. Due to the nature of probability, this value can only be in the interval <code>(0; 1)</code>. We will use the values 0.9, 0.95, and 0.99 to maintain scientific rigor and ease of interpretation, opposite to the [[https://en.wikipedia.org/wiki/Statistical_significance significance level]].
=== Assumptions ===
;Assumption 1
: All <code>k</code> centrifuges work continuously and simultaneously for each <code>n</code> cycle with the same <code>τ</code>
;Assumption 2
: There are always enough resources to convert [[uranium-235]] into reactor fuel *
;Assumption 3
: The conversion of [[uranium-235]] into fuel for reactors can be neglected **
;Assumption 4 (for long-term play)
: The game can be played indefinitely (technically assumed for practical use, average playtime suffices)
<nowiki>*</nowiki> For instance, base game [[uranium-238]] has a very high probability (<code>0.993</code>), ensuring a sufficient supply for [[uranium fuel cell]] production.
<nowiki>**</nowiki> For instance, converting base game [[uranium-235]] into 10 [[uranium fuel cell]]s takes <code>12 sec</code>, but the [[reactor]] cycle is <code>200 sec</code> - a negligible time difference, easily managed by building more [[assembler]]s.
=== Exact Approach ===
The main question we aim to answer:
<blockquote>
How many <code>k</code> [[centrifuge]]s are needed during <code>n</code> cycles of [[centrifuge]] production to sustain a [[reactor]] working continuously for <code>T</code> time with probability <code>&alpha;</code>?
</blockquote>
Assume each [[centrifuge]] in each production cycle is represented by a discrete random variable '''''X<sub>ij</sub>''''', where '''''i''''' is the [[centrifuge]] number and '''''j''''' is the production cycle number. If '''''X<sub>ij</sub>''''' produces [[uranium-235]], its value is <code>X<sub>ij</sub>=1</code>. If not, <code>X<sub>ij</sub>=0</code>. Thus, <code> P(X<sub>ij</sub> = 1) = 0.007</code> and <code> P(X<sub>ij</sub> = 0) = 0.993</code> (This is a [[https://en.wikipedia.org/wiki/Bernoulli_distribution Bernoulli distribution]]).
To find out how long [[centrifuge]]s should produce [[uranium-235]] to sustain a [[reactor]]:
From one side, the [[reactor]] has <code>T/t</code> working cycles, needing a total of <code>Tc/t</code> [[uranium-235]] to run continuously for time '''''T'''''. Let's denote this value as '''''u''''': <code>u=Tc/t</code>.
From the other side, the [[centrifuge]]s work continuously for '''''n''''' cycles during time '''''T''''': <code>T = n*τ</code>.
Combining these, our required quantity of '''''u''''' is <code>u=n*τ*c/t</code> for the entire [[reactor]] operation time '''''T'''''.
Consider the simple case where we need '''''n''''' cycles and '''''k''''' [[centrifuge]]s to produce exactly '''''u''''' [[uranium-235]] (technically, we need ceil of '''''u''''', but it won't be practically useful and doesn't detract generality of our calculations). Each '''''X<sub>ij</sub>''''' is a Bernoulli trial, so <code>X<sub>all</sub> = ∑<sub>i</sub><sup>n</sup>∑<sub>j</sub><sup>k</sup>X<sub>ij</sub></code> is the total production, distributed as a [[https://en.wikipedia.org/wiki/Binomial_distribution binomial distribution]] <code>X<sub>all</sub> ~ B(nk, p)</code>. This gives the probability of producing exactly '''''u''''' [[uranium-235]] in '''''n''''' cycles with '''''k''''' [[centrifuge]]s:
<blockquote>
<code>P(X<sub>all</sub>=u) = C<sub>nk</sub><sup>u</sup> * p<sup>u</sup> * (1-p)<sup>nk-u</sup></code>
</blockquote>
Practically, we are interested in producing at least '''''u''''' [[uranium-235]], corresponding to <code>P(X<sub>all</sub> ≥ u)</code>. This can be calculated as <code>P(X<sub>all</sub> ≥ u) = 1 - P(X<sub>all</sub> < u)</code>, where <code>P(X<sub>all</sub> < u)</code> is the [[https://en.wikipedia.org/wiki/Binomial_distribution#:~=0.059535.-,Cumulative%20distribution%20function,-%5Bedit%5D cumulative distribution function]]:
<blockquote>
<code>P(X<sub>all</sub>≥u) ≥ &alpha;</code> ⇒ <code>1 - &sum;<sub>m</sub><sup>u-1</sup>P(X<sub>all</sub>=m) ≥ &alpha;</code> ⇒ {using results from paragraph 1} ⇒ <code>&sum;<sub>m</sub><sup>n*&tau;*c/t-1</sup>C<sub>nk</sub><sup>m</sup> * p<sup>m</sup> * (1-p)<sup>nk-m</sup> ≤ 1 - &alpha;</code>
</blockquote>
So, we now have a formula to determine the number of '''''k''''' [[centrifuge]]s needed to sustain a [[reactor]] with consumption '''''c''''' during '''''n''''' cycles. This formula is general and not always practical for long-term play due to the high values of '''''n''''' and '''''u'''''. It is best suited for short-term or speedrun calculations.
==== Example Base Game Short-Term ====
To illustrate, let's determine how many [[centrifuge]]s are needed to sustain 1 [[reactor]] with 0.99 confidence without modules. It's a common mistake to think that 0.007 is the production speed, but it represents the probability of producing [[uranium-235]] per cycle. Suppose we want a reactor to work for 2000 [[centrifuge]] production cycles (about 6.7 hours of reactor operation) with 1 [[centrifuge]]:
{| class="wikitable"
! n !! tau !! c !! t !! m !! p !! u !! ceil(u-1) !! alpha !! 1-alpha !!  !! k !! C(nk, m) !! P(m) !! &sum;<sub>m</sub>P
|-
| 2000 || 12 || 0,1 || 200 || 0 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 1 || 7,92E-07 || 7,92E-07
|-
| 2000 || 12 || 0,1 || 200 || 1 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 2000 || 1,12E-05 || 1,20E-05
|-
| 2000 || 12 || 0,1 || 200 || 2 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 2,0E+06 || 7,86E-05 || 9,06E-05
|-
| 2000 || 12 || 0,1 || 200 || 3 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 1,3E+09 || 0,0004 || 0,0005
|-
| 2000 || 12 || 0,1 || 200 || 4 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 6,6E+11 || 0,0013 || 0,0018
|-
| 2000 || 12 || 0,1 || 200 || 5 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 2,7E+14 || 0,0037 || 0,0054
|-
| 2000 || 12 || 0,1 || 200 || 6 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 8,8E+16 || 0,0086 || 0,0140
|-
| 2000 || 12 || 0,1 || 200 || 7 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 2,5E+19 || 0,0172 || 0,0312
|-
| 2000 || 12 || 0,1 || 200 || 8 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 6,3E+21 || 0,0302 || 0,0302
|-
| 2000 || 12 || 0,1 || 200 || 9 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 1,4E+24 || 0,0472 || 0,0472
|-
| 2000 || 12 || 0,1 || 200 || 10 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 2,8E+26 || 0,0662 || 0,1133
|-
| 2000 || 12 || 0,1 || 200 || 11 || 0,007 || 12 || 11 || 0,99 || <span style="color: red">0,01</span> ||  || 1 || 5,0E+28 || 0,0844 || <span style="color: red">0,1977</span>
|-
| 2000 || 12 || 0,1 || 200 || 12 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 8,3E+30 || 0,0986 ||
|-
| 2000 || 12 || 0,1 || 200 || 13 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 1 || 1,3E+33 || 0,1063 ||
|}
As you can see, we need to produce 12 '''''u''''' for 2000 cycles. According to our formula, we need to sum up all the probabilities of producing less than 12 [[uranium-235]] and compare this with our desired confidence level:
<code>∑<sub>m</sub><sup>2000120.1/200-1</sup>C<sub>20001</sub><sup>m</sup> * 0.007<sup>m</sup> * (1-0.007)<sup>20001-m</sup> = 7.92E-07 + 1.12E-05 + 7.86E-05 + 0.0004 + 0.0013 + 0.0037 + 0.0086 + 0.0172 + 0.0302 + 0.0472 + 0.0662 + 0.0844 = 0.1977</code>
We can interpret this value as the "Probability of producing less than the required amount of [[uranium-235]] for 2000 [[centrifuge]] cycles." This probability should be less than <code>1-α = 0.01</code>. However, since it is not, we cannot be sure that 1 [[centrifuge]] will produce enough [[uranium-235]] to sustain [[reactor]] operation during 2000 cycles. Therefore, we need to increase the number of [[centrifuge]]s.
{| class="wikitable"
! n !! tau !! c !! t !! m !! p !! u !! ceil(u-1) !! alpha !! 1-alpha !!  !! k !! C(nk, m) !! P(x=m) !! sum(P(m<=ceil(u-1)))
|-
| 2000 || 12 || 0,1 || 200 || 0 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 1 || 6,3E-13 || 6,27E-13
|-
| 2000 || 12 || 0,1 || 200 || 1 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 4000 || 1,8E-11 || 1,83E-11
|-
| 2000 || 12 || 0,1 || 200 || 2 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 8,0E+06 || 2,5E-10 || 2,67E-10
|-
| 2000 || 12 || 0,1 || 200 || 3 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 1,1E+10 || 2,3E-09 || 2,61E-09
|-
| 2000 || 12 || 0,1 || 200 || 4 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 1,1E+13 || 1,6E-08 || 1,91E-08
|-
| 2000 || 12 || 0,1 || 200 || 5 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 8,5E+15 || 9,3E-08 || 1,12E-07
|-
| 2000 || 12 || 0,1 || 200 || 6 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 5,7E+18 || 4,4E-07 || 5,48E-07
|-
| 2000 || 12 || 0,1 || 200 || 7 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 3,2E+21 || 1,8E-06 || 2,30E-06
|-
| 2000 || 12 || 0,1 || 200 || 8 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 1,6E+24 || 6,2E-06 || 6,17E-06
|-
| 2000 || 12 || 0,1 || 200 || 9 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 7,2E+26 || 1,9E-05 || 1,93E-05
|-
| 2000 || 12 || 0,1 || 200 || 10 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 2,9E+29 || 0,0001 || 0,0001
|-
| 2000 || 12 || 0,1 || 200 || 11 || 0,007 || 12 || 11 || 0,99 || <span style="color: red">0,01</span> ||  || 2 || 1,0E+32 || 0,0001 || <span style="color: red">0,0002</span>
|-
| 2000 || 12 || 0,1 || 200 || 12 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 3,4E+34 || 0,0003 ||
|-
| 2000 || 12 || 0,1 || 200 || 13 || 0,007 || 12 || 11 || 0,99 || 0,01 ||  || 2 || 1,1E+37 || 0,0007 ||
|}
And now you can see that our condition is fulfilled. Therefore, we can say that without any [[module]]s or [[Kovarex enrichment process]], we can achieve the stability of 1 [[reactor]] during 2000 [[centrifuge]] cycles using at least 2 [[centrifuge]]s! However, does this imply that we need 4 [[centrifuge]]s for 2 [[reactor]]s, 6 [[centrifuge]]s for 3 [[reactor]]s, and so on? I encourage you to think about this question yourself. You may be surprised by the results when applying the approach outlined above.
=== Distribution Approximation ===
If you plan to play for an extended period while your [[reactor]] setup is operating, we can consider <code>n→∞</code>. In this case, since our [[centrifuge]]s' overall production <code>X<sub>all</sub> ~ B(nk, p)</code>, we can use the [[https://en.wikipedia.org/wiki/De_Moivre%E2%80%93Laplace_theorem De Moivre-Laplace theorem]]. In short, it states that if our random variable follows a binomial distribution with a large number of trials, then <code>(X<sub>all</sub>-nkp)/sqrt(nkp(1-p)) ~ N(0, 1)</code>. Thus, to estimate <code>P(X<sub>all</sub> ≥ u)</code>, we simplify it to:
<code>P((X<sub>all</sub>-nkp)/sqrt(nkp(1-p))) ≥ (u-nkp)/sqrt(nkp(1-p)) ≥ α</code> ⇒ <code>{1-P<sub>dist</sub> ≥ α}</code> ⇒ <code>(nkp - u)/sqrt(nkp(1-p)) ≥ ɸ<sup>-1</sup>(1-α)</code>, where ɸ<sup>-1</sup> is the inverse standard normal distribution, found using the [[https://en.wikipedia.org/wiki/Standard_normal_table standard normal table]].
Let's denote <code>ɸ<sup>-1</sup>(α)</code> as λ. For each confidence level α, we can calculate it from the standard normal table:
<code>λ<sub>0.99</sub>=2.3263</code>, <code>λ<sub>0.95</sub>=1.6449</code>, <code>λ<sub>0.9</sub>=1.2816</code>.
Now we need to solve the equation above for the variable '''''k''''':
<code>(nkp-u)/sqrt(nkp(1-p)) ≥ λ</code> ⇒ <code>(nkp-nτc/t)/sqrt(nkp(1-p)) ≥ λ</code>.
After some basic arithmetic operations and solving the quadratic equation, you can determine the value of '''''k''''':
<blockquote>
<code>k ≥ (χ<sub>1</sub> + sqrt(χ<sub>1</sub><sup>2</sup> + 4χ<sub>0</sub>χ<sub>2</sub>))/(2*χ<sub>0</sub>)</code>
where
<code>χ<sub>0</sub> = np</code>,
<code>χ<sub>1</sub> = &lambda;*sqrt(np(1-p))</code>,
<code>χ<sub>2</sub> = n&tau;c/t</code>,
<code>&lambda;</code> is the inverse standard normal distribution (popular values: 2.3263; 1.6449; 1.2816 for corresponding &alpha; values: 0.99; 0.95; 0.9).
</blockquote>
This formula can be calculated for very large '''''n'''''. That's why it's best choice for planning average playtime of experienced player.
=== Infinite Approximation ===
If we consider the formula for '''''k''''' [[centrifuge]]s above, we can derive an analytical form for <code>n→∞</code> by substituting all '''''χ''''' into the formula and finding the limit:
<blockquote>
<code>lim<sub>n→∞</sub>(χ<sub>1</sub> + sqrt(χ<sub>1</sub><sup>2</sup> + 4χ<sub>0</sub>χ<sub>2</sub>))/(2*χ<sub>0</sub>)</code> ⇒ {substitute all '''''χ'''''} ⇒ <code>k ≥ &tau;*c/(p*t)</code>
</blockquote>
Thus, to determine the minimal number of [[centrifuge]]s needed to achieve [[reactor]] stability, we use the final formula:
<blockquote>
<code>k = ⌈&tau;*c/(p*t)⌉, where ⌈x⌉ is the ceiling function</code>
</blockquote>
Notice that this formula does not depend on '''''λ'''''. The main reason is that when <code>n</code> is infinitely large, the cumulative probability tends to 1, and at infinity, we can be 100% sure that this number of [[centrifuge]]s will sustain our [[reactor]]'s stability forever!
This formula provides the minimal number of [[centrifuge]]s needed to sustain [[reactor]] stability indefinitely. To determine this number, we need only 4 parameters.


== History ==
== History ==

Latest revision as of 08:33, 30 August 2024

Uranium processing.png
Uranium processing

Recipe

Time.png
12
+
Uranium ore.png
10
Uranium-235.png
0.007
+
Uranium-238.png
0.993

Total raw

Time.png
12
+
Uranium ore.png
10

Prototype type

recipe

Internal name

uranium-processing

Required technologies

Uranium processing (research).png

Produced by

Centrifuge.png

Uranium processing is the only way to use uranium ore and the first source of uranium-235 and uranium-238 that is available to the player. The process has a 99.3% chance to produce 1 uranium-238 and a 0.7% chance to produce 1 uranium-235.

Initiating the Kovarex enrichment process

Unlike all other crafting processes in the game, uranium processing creates U-235 and U-238 based on probability, rather than in guaranteed deterministic amounts. Additionally, the Kovarex process requires a bulk stockpile of 40 units of the rare uranium-235 isotope to initiate in a single centrifuge. Therefore, planning uranium mining and processing for the Kovarex process takes certain considerations not encountered elsewhere in the game.

Expected value

As U-235 is created in centrifuges running uranium processing with p = 0.007, the expected number of processing cycles to gain a single unit is E(1,p) = 1 ÷ p = ~143. The expected number of cycles to gain 40 units is then E(40,p) = 40 × E(1,p) = ~5,714 cycles. Note that running this many cycles requires mining 57,140 uranium ore, a non-trivial task.

However, given the nature of probability, it is impossible to guarantee that 40 units will be reached in this number of cycles; as a matter of fact, some mathematics yields that the probability that 5,714 cycles will produce at least 40 units of U-235 is only about 52%.

Confidence levels

Probability of reaching 40 U-235 from uranium processing by number of cycles (click to enlarge)

Some further mathematics enables one to calculate the number of uranium processing cycles one must run to obtain at least 40 units of U-235 (given p = 0.007) with a given level of confidence (probability of achieving the set goal).

As discussed in the previous section, the level of confidence corresponding to 5,714 cycles (for which the expected value of U-235 gained is 40) happens to be about 52% (or, in other words, with this number of cycles, one can expect to fail to reach 40 U-235 a bit less than once out of every 2 tries).

Some further example confidence levels that may be of interest to players are given in the table below. Consult the chart to the right for custom values.

Confidence level Failures Cycles required
10% 9 out of 10 4,595
50% 1 out of 2 5,667
90% 1 out of 10 6,894
95% 1 out of 20 7,272
99% 1 out of 100 8,015

History

See also