Friday, June 28, 2019

Fleet Combat Dynamics II

"All models are wrong, but some are useful."
- George Box

In my post in March 2018, I presented a continuous dynamical model that enabled analysis of homogeneous fleet combat in aggregate terms.  By modeling only the damage rates, effective hit points and number of ships in each fleet, I showed that a pair of separable differential equations could model dynamics of ship losses with time.  

Using this model as a starting point, I showed a couple basic things.  First, I showed that the number of ships in your fleet has a squared effect on fight outcomes.  Stated simply, bringing just 10% more people is roughly equivalent to fitting every ship with 20% more hit points per ship.  Second, I showed that at any given time during a fight, comparing the fractional rate of ship loss for two fleets is equivalent to comparing the figure of merit for each fleet. 

In this short post I want to show a good example of how the dynamic model behaves for a real fight, and suggest extending the interpretation of this dynamical model for fleet combat to help us diagnose the outcome of another engagement.

I'm including footage of the fight in the link below.  Thanks to BigChols and the Chols Media Group for providing the community with videos of important PVP engagements!

I highlight this fight because it has many of the properties that I relied upon in the fleet combat dynamics model that I shared in March 2018.  That is, this fight involves two fleets that are largely homogeneous and have similar numbers, when counting their damage dealing ships.  

Even more intriguing is that the damage dealing ships are of the same type in both fleets which is a good case to study with our dynamic model.  I will first make the assumption that both fleets are fitted and trained identically and that hit-points and damage rate are the same for both ship types, i.e. $HP_A = HP_B$, as well as, $DPS_A = DPS_B$.  This makes the assumption that both fleets have the same fitting, use the same ammo type, and on average, have the same level of skill point training associated with the fittings of their ships.  These assumptions are not far fetched, and as you will see below, are mostly correct.  

The battle report for the fight is here (ZMV9-A battle report) and I summarized the initial numbers involved below:

Ship Type \ Fleet  Fleet A
(Black Legion) 
Fleet B (Test Alliance)
DPS Muninns 104 94
Logistics Loki 5 2
 Shield Logistics 6 0

Figure 1 below shows the time evolution of the number of damage dealing ships surviving with time.  The plots in this post show time zero aligned with 15.**.  Also, note that I only plot the solution to the dynamic model until the 480 second point, or eight minutes.  From the footage, my understanding is that Test Alliance Munnins had disengaged at around 400 seconds and were leaving grid after that point, so they were not applying damage as effectively.  Futher model calculation wouldn't be relevant to understanding the outcome.  

Figure 1: Damage dealers (Munnins) remaining alive over time. 

Clearly Fleet A had superior damage-dealer numbers as well as logistics.  The results aren't surprising in the number of ships destroyed when Fleet B decides to leave the field.  

If I ignore the effect of logistics, and whatever fitting and ammo differences there are, can our simple statistical dynamical model capture the overall outcome of the engagement?  If I only modeled the damage dealer behavior and ignored everything else, how well would our simple model predict the outcome of this fight?

Based on the fleet composition information and basic information from ship fitting I use the following parameters in the dynamic model:

Ship Type \ Fleet  Fleet A
(Black Legion) 
Fleet B  
(Test Alliance)
EHP 75000 75000
Effective DPS 120 120

Note these are just estimated from the usual fittings, resists and ammo used by artillery Munnins.  $T_{mutual}$ is estimated at ~610 sec.

So, I want to compare the model assuming that both fleets are equally coordinated and using the same ammunition.  Recall that the closed form of the dynamical model estimates the time evolution of the surviving numbers, as the number $N_a(t)$ and $N_b(t)$, 
$N_a(t) = N_A cosh(t/T_{mutual}) - N_B \sqrt{ \frac{DPS_B HP_B}{ DPS_A HP_A} } sinh(t/T_{mutual})$
$N_b(t) = N_B cosh(t/T_{mutual}) - N_A \sqrt{ \frac{DPS_A HP_A}{ DPS_B HP_B} } sinh(t/T_{mutual})$ 

You can review the derivation of the model here.

Plotting these equations in Figure 2, I compare the remaining number of damage dealers during the eight minutes of the engagement.  There is surprising agreement between the model and the actual loss rates for both fleets.  Looking a bit more closely, the loss rates are almost constant, indicating that the fight lasted less than a single engagement time-constant, $T_{mutual}$.  This is quite different than the example I included in my previous post, because in this case there are no external damage dealers confusing the assumptions of the model. 

Note also that the loss rate of the BL fleet appears to drop to zero after the six minute mark (360 seconds), while Test fleet losses continue to mount for another two minutes.  

Figure 2: Dynamical model playback with $HP_A = HP_B = 75000$. 
Qualitative agreement is surprisingly good considering the complexity of this fight
and the presence of a small number of logistics.   

Looking more closely at Figure 2, there are a couple things to mention.  First, you'll notice that I put the zero start time after some of the Fleet B ships had already been lost.  This is because in the first 60 seconds, Fleet B was shooting logistics ships and Lokis.  Considering the zero-time I'm using in this plot is helping me to ignore logistics, because some of these have been eliminated.  Second, there are some notes that might be worth adding on damage application and command links [bunch more to be added later].  

Having mentioned these factors, another thing that can contribute to difference between the model and the fight outcome is that Test Muninns appear to be loading artillery with Titanium Sabot, while Black Legion is loaded with Fusion.  Seeing as the fleets are at the same aggregate distance from one another, there does not appear to be any reason to use longer range ammunition.   

Executive Summary

  • The strength of your doctrine fleet scales with the square of the number of damage dealing ships, i.e. $O(N^2)$.  Another way to think of this is that once you have lost 30% of your combat ships (or 70% of damage dealers are remaining) you have lost more than half of your combat capability.
  • Ammunition choice, resists and target order have an effect on fight outcomes.  Don't let that part of the fleet doctrine and command go unexamined or unpracticed.

Friday, March 16, 2018

Dynamics of Homogeneous Fleets in Combat

"No one wins. One side just loses more slowly."
- Roland Pryzbylewski - The Wire

When I published my first set of notes on EVE Online in 2010, I titled the work, "Mathematics of EVE Online." Seeing as my first complete chapter was exclusively on the physics of object motion in the game, it should not have surprised me that one of the criticisms of the work was that I should have named it, "Applied Physics in EVE Online." I had not, at that time, emphasized that there are many other topics to study that have nothing to do with motion physics whatsoever.

The derivation work presented in this note began eight years ago.  It represents a portion of the tactical decision abstracts that were shared internally with Origin. and BPI at that time. It is my hope that the wider community will also find them of interest today, in spite of the significant changes that have taken place in the game.

Now, we all know that fleet battles are extremely complex. It is impossible to determine the outcome of any conflict beforehand. All the same, I want to ask some questions about whether it is possible to identify statistics for outcomes of engagements.  If we make some basic assumptions based on observations, what range of outcomes can we determine for fleet action? What follows in this and subsequent posts is intended to clarify some of the components that go into combat decisions and dynamics, such as:
  • Should we engage this hostile force or not?
  • At what point should our fleet attempt to flee? 
  • Should we prioritize target A or target B?
Readers should be forewarned, however, that the value is not in the instructive nature of the results in these articles, but instead in formalizing the thinking process that is at work in current fleet decision making.  My intention here is more narrow than you might think and I include a note about this below.  

I will add notes to the Executive Summary page to help with readers who are pressed for time. 


This note presents a simple analysis to help inform decision making for fleets based on understanding the dynamics of combat between two groups of ships. As it stands, the assumptions in this note are very narrowly defined. Instead of seeing this as a negative, see it for what it is — narrowly instructive, but broadly analytical. Any model is a product of a set of assumptions, and by making different assumptions you will get different results.  This model represents a starting point for understanding the statistical dynamics of fleet combat.

These notes are not intended to prove or disprove the balance of strategies, or specific ship-types.  They are not intended as a training guide.  Quite the contrary, all I am trying to do here provide fleet commanders with a tool that can be used to analyze their performance in fleet combat situations and help them clarify which decisions they might change in the future.  If you disagree with the approach or my interpretation of them I encourage you to make an effort to quantify  your decision making and share it with the EVE community.  

With some additional work, some of which I may include in a later post, this approach can take into account averaged locking delay, active tanking and even some electronic countermeasure effects. It can not, however, take into account speed tanking or extensive electronic warfare setups.  Thus it is not at all suitable for analyzing nano-gangs, teams of stealth bombers, electronic warfare-heavy groups, or fleets where capitals are changing state, i.e. siege/bastion/triage-modes, or alpha-maximized fleets whose volley damage is so high that ships are destroyed significantly faster than it takes to lock them.  

Finally, readers who are familiar with dynamic population models might characterize the model I present below as a case of "mutual-predators without replacement."  Not a great deal has been written about this case because, as you will see, one team all dies and the fun is over.  From a strategy perspective, I think there is more to be said about this model, so stay tuned.

On with some assumptions... 

Time And Loss

In fleet combat, as in your life, time is the only thing that you can't go get more of.  If you participate in enough large fleet combat you can observe some of the factors that contribute to the time required to do battle with an enemy fleet.  When the fleet commander gives a command to destroy a ship, a few things have to happen.  First, pilots have to lock the target, then pilots have to activate weapons, then missiles or drones (if they are being used) have to reach the target, and then the damage from attackers has to fully deplete the hit-points of the target.  I will group the hit-point groups (shields, armor and hull) into an effective hit-point number, $HP_{target}$, that also accounts for resist ratios.  In this way, I calculate that time to destroy a ship takes the form,
$\large T_{kill} = T_{lock} + T_{flight time} + \frac{HP_{target}}{DPS - RPS}$

That is, a target ship has $HP_{target}$ hit points and is experiencing a repair rate (either external or internal) of RPS effective hit points repaired per second. This form assumes that you have scaled the repair rate by the resists so that incoming damage can be compared directly to find the total time until the ship is destroyed.  This is not as simple as it looks but assume that we have these numbers. It also assumes that the repair rate is less than the incoming damage rate, which often depends on target because repair effectiveness is critically linked with resist ratio. For this note, assume that $DPS_{eff} = DPS - RPS$ and that hostile $DPS$ is greater than logistics repair rates, or $DPS > RPS$. 

The sum also includes the time it takes to lock the target, or $T_{lock}$.  The locking time term, $T_{lock}$, is a simple function of your scan resolution and the target's signature radius.  Disciplined fleets prelock a secondary target while they shoot at the primary, making this term small or irrelevant. If missiles or drones are your sole damage dealing mechanism, you now wait for them to arrive at the target, or $T_{flight time}$.  Assume turret damage has zero flight time.  

I have included a note about other nonidealities with this rough model in Appendix A at the end of this note, including the 'Dumbo Factor', which can be important to consider. 

The Assumptions

In this article I will consider the case where two fleets, each comprised of homogeneous ships, are in combat.  Imagine two fleets:  The first fleet having $N_A$ total ships at the start of the combat, each ship having effective $HP_A$ effective hit-points and delivering effective damage-per-second of $DPS_A$; The second fleet starts with $N_B$ ships, each with effective $HP_B$ hit-points and delivering effective $DPS_B$ damage-per-second.  I will also assume that neither fleet targets enemy logistics and repair effects are are effectively constant for both fleets.

Now the rate of damage may not be constant, but assume that we chew through the total hit points at a fixed effective target armor, which takes $\frac{HP_{target}}{DPS_{eff}}$ time to destroy the target.  If you are imagining that the damage rate and hit-points are more complicated than that, of course you are correct.  For this article, however, I am using an effective damage rate that takes into account all of the mitigating factors and is effectively computing the time that the target survives while under assault.

In light of the small contributions from the locking time and flight time, I want to make the assumption going forward that $T_{kill}$ is dominated by the amount of time needed for the effective damage per second rate to chew through each target's effective hit-points net of repair activities.  That is,
$\large T_{kill} \simeq \frac{HP_{eff}}{DPS_{eff}}$

This type of fleet combat situation is quite common.  In this fight, the fleet commander has a wide mix of cruiser-class damage dealers.  In spite of this, there is reasonably small variability in the kill-time of targets.  I recorded the kill times for Auguror Navy Issue cruisers starting from the time the target is called to the time it is destroyed in Figure 1, below.  Because of the way I am collecting this data, you will see several outliers, but I include these because it would be misleading not to.  In those outlier cases, however, there are other targets called as priority in the interim and these times also include at least one reload cycle.  If you review the video you'll see that these outliers don't block the progress of the fight.  You can see that under normal conditions kill-time duration is clustered around a statistical mean near seven seconds.

Figure 1: A histogram of the times from target called to target destruction, or kill-time, is shown.  Fight was from GE-8JV example in the text.  Data includes target recalls, damage splitting and reloading cycles in a complex fight scenario.

These data are relevant to a discussion about statistical fleet behavior, but that is not my goal here.  Instead, I want to ask questions given that fleets behave statistically.  That is, what can we conclude about the dynamics and outcomes of these kinds of pitched battles?  If every pilot behaves exactly as the fleet commander orders them, targeting the primary with all their damage and prelocking secondary targets, which fleet will be left standing?  Do the answers to these questions suggest decision points where fleet commanders should choose to disengage, or change positioning to improve effective damage?  

Thinking about time in fights can be powerful.  Before I derive dynamics, a couple more notes:  First, fights where locking time dominates combat duration are usually between groups who are significantly size mismatched, or where target calling is inconsistent. I am not considering this case here.  

Second, in an upcoming article, I will analyze some combat decisions between heterogeneous groups and review some basic results about decision making in that context.  In that post, I will show how having complete information about the properties of a heterogeneous enemy fleet allows target ordering that can optimize for several outcomes.  

Finally, the concept of effective damage-per-second and effective hit-points are intended to avoid dependence on the obvious challenge that every fleet commander faces in trying to position their fleet for maximum effectiveness, choose ammunition for best effect, choose hardeners to best resist incoming damage and fitting for mobility sufficient to maintain a desirable engagement range.  These are all critically important, but I will assume here that whatever our fleet commanders have done remains constant in this regard, maximizing their total performance given their degrees of freedom.  This is where I think generalization to specific situations will break down, but in return I promise to offer insight into pitched battle dynamics. 

Without further delay, lets dive in to the model. 

Deriving The Model

First, I write the number of ships left alive at time, $t$, by subtracting the number of ships destroyed at time $t$ from the initial number of ships in each fleet. Note that $N_a(t)$ and $N_b(t)$ are functions that represent the number of ships in each fleet at time $t$, and $N_A$ and $N_B$ are the initial number of ships in each fleet.
$\large N_a(t) = N_A - \int_0^t \frac{N_b(t) DPS_B(t)}{HP_A} dt$
$\large N_b(t) = N_B - \int_0^t \frac{N_a(t) DPS_A(t)}{HP_B} dt$     Eq. 2–1

Writing this as a rate of change for the number of ships over time,
$\large \frac{dN_a(t)}{dt} = - \frac{DPS_B}{HP_A} N_b(t)$
$\large \frac{dN_b(t)}{dt} = - \frac{DPS_A}{HP_B} N_a(t)$     Eq.2–2

Note that Eq. 2–2 is just a restatement of Equation 2–1, that the rate of change of ships in a fleet depends on the time that it takes for a ship in the enemy fleet to destroy one ship multiplied by the number of hostiles that are shooting at you. Thus, Eq. 2
2 should make intuitive sense to all readers.   

Now, let's write the equations only in terms of ships in their respective fleets,
$\large \frac{d^2 N_a(t)}{dt^2} = \frac{DPS_A DPS_B}{HP_A HP_B} N_a(t)$
$\large \frac{d^2 N_b(t)}{dt^2} = \frac{DPS_A DPS_B}{HP_A HP_B} N_b(t)$     Eq.2–3

I like these equations because they force us to consider what is really happening in a fleet fight. The first derivative equation (Eq. 2–2) was negative because the number of ships in your own fleet is decreasing and at a rate that is proportional to how many hostiles are shooting at you as well as their damage capability

The second derivative (Eq. 2–3) gets to another truth about fleet combat the rate at which you decrease the harm to your own fleet depends on how quickly you are destroying the enemy's ability to destroy you.  This rate does not depend on the number of hostiles.

Another thing worth noticing about Eq.2–3 is that the coefficients are symmetric between both changes in ship-number rates. These take the form of the time-to-kill for a one-on-one match-up between ships in the respective fleets. That is,
$T_{A \, kills \, B} = \frac{HP_B}{DPS_A}$
$T_{B \, kills \, A} = \frac{HP_A}{DPS_B}$

The geometric mean of these times will keep popping up throughout the solution to this system. I will simplify the writing by calling this quantity $T_{mutual}$,
$\large T_{mutual} = \sqrt{ T_{A \, kills \, B} \, T_{B \, kills \, A} } = \sqrt{\frac{HP_B \, HP_A}{DPS_A \, DPS_B}}$.

This time-constant, $T_{mutual}$, is a characteristic duration of dynamics of the combat between these fleets, corresponding to the time when the effect of one fleet's damage overwhelms the damage from the other fleet, meaning that the loss rates of the two groups begin to diverge. If you are in sustained engagement for a duration of $T_{mutual}$ or longer, the fight will be well decided. 

Now that I have written each fleet's ship numbers as a rate in one equation, the method of solution is very easy. I solve for the actual ship numbers versus time, $N_a(t)$ and $N_b(t)$. I've skipped some details here but we know the form of the solution, because differential equations of this type are so common in nature.

Mixing in a boundary condition that the initial number of ships and the initial rate of ship destruction, I find the time domain equation for the number of ships during the battle,
$\large N_a(t) = N_A cosh(t/T_{mutual}) - N_B \sqrt{ \frac{DPS_B HP_B}{ DPS_A HP_A} } sinh(t/T_{mutual})$
$\large N_b(t) = N_B cosh(t/T_{mutual}) - N_A \sqrt{ \frac{DPS_A HP_A}{ DPS_B HP_B} } sinh(t/T_{mutual})$         Eq. 2–4

I've written this as a number of ships remaining alive at time $t$.  Equations Eq.2
–4 is a competition between hyperbolic functions representing the two forces. There is something deeper going on here, but it will take a bit more work to tease apart what is happening. In the meanwhile, lets look at some example dynamics of this model by plotting the number of ships over time.

Note that the solution in Eq.2–4 is valid for $t > 0$ and $N_a(t) > 0$ and $N_b(t) > 0$ as well as the initial conditions where the fleet properties are not identical, an infinite solution artifact arising from modeling the number of ships as a continuous variable.

Example Dynamics

I show a few examples of this fleet combat dynamic model below.  These examples are chosen to illustrate how these solutions behave all the way to when one of the fleets is completely eliminated even though most fleet commanders will make an effort to leave the field long before that time.  I've also constructed the example to show some basic trade offs related to number of ships versus quality of ships.  In Figure 2 below, I show the model dynamics for two fleets with equal numbers but where Fleet A has only 10% higher DPS per ship than Fleet B.  This would be a close fight, and Fleet A wins the fight, eliminating hostiles but with only 30% of their ships surviving.  

Figure 2: Example of two fleets with identical starting numbers but differing ship quality is shown.  Fleet properties: $N_A = N_B = 100$, $HP_A = HP_B = 30000$, $DPS_A = 110$ $DPS_B = 100$, $T_{mutual} = 286 \, seconds$.

The second example in Figure 3, below, changes the conditions slightly, showing the model dynamics for a case where Fleet A this time brings 10% more pilots than Fleet B, but the ship performance in both fleets, DPS and HP, are now identical.   Note that the surviving number of Fleet A ships is now 42% of the starting number.  I hope to explain the intuition behind this difference in the Discussion section below.  

Figure 3: Example of two fleets with identical ship quality but differing initial numbers is shown.  Fleet properties: $N_A = 110$, $N_B = 100$, $HP_A = HP_B = 30000$, $DPS_A = DPS_B = 100$, $T_{mutual} = 300 \, seconds$.

In both of these examples I have shown 'Time' on the x-axis, per the model, which does not include locking time or time-of-flight for missile or drone damage so actual engagement duration may be different. 

I'll share one more example where far superior ships can defeat superior numbers in Figure 4.  I constructed this case as a hypothetical situation where heavy assault cruisers might best some Tech I cruisers but the exact numbers are not critical to demonstrating the qualitative nature of dynamics for this model. 

Figure 4: Example of two fleets with differing ship numbers and quality is shown.  Fleet properties: $N_A = 50$, $N_B = 100$, $HP_A = 70000$, $HP_B = 30000$, $DPS_A = 200$, $DPS_B = 100$, $T_{mutual} = 324 \, seconds$.

I have shown engagements examples above that result in complete destruction of one of the two fleets. Note that in all of these examples, as the losing fleet's numbers diminish to zero, the rate of loss for the winning fleet levels off, reflecting the withering incoming damage from the vanquished.

With these pictures in mind, I'd like to derive a few other results that arise from this viewpoint.  The most obvious question we want to be able to answer being, "given the numbers of ships and their properties, which group will be eliminated first?"  Of course, shrewd fleet commanders will make an effort to escape long before complete destruction but we can use the answer to this question to guide that decision as well.  After all, holding the field after the fight advantages the survivors with the ability to loot, so holding the field by complete destruction of the opponent, or forcing them to flee will both be treated as desirable outcomes. 

A Figure of Merit

I intentionally cropped the example graphs above at the point where the losing fleet's ship numbers dropped to zero. When does this happen? If I write the solution for $N_*(t) = 0$, solving for the time when the opposing force has zero ships, I find,
$t_{death,B} = T_{mutual} \, tanh^{-1} \left ( \frac{N_B}{N_A} \sqrt{ \frac{DPS_B HP_B}{DPS_A HP_A} } \right )$
$t_{death,A} = T_{mutual} \, tanh^{-1} \left ( \frac{N_A}{N_B} \sqrt{ \frac{DPS_A HP_A}{DPS_B HP_B} } \right )$         Eq 2–5

The defeated party in the engagement will be the one who has the lesser of these two death times.  Just remember that $tanh^{-1}(x)$ is only real for $|x|$ less than 1.  So, the fleet that is defeated will be the fleet having the lower $N \times \sqrt{DPS \times HP}$ product.

This is an important result because the benefit of ship numbers and ship quality are not the same. I hinted at this in my dynamics examples above. Because the $N$ term has squared ratio to the $DPS \times HP$ product, bringing 10% more ships is approximately as effective as bringing 20% more damage per ship.  Or, bringing 10% more ships is approximately as effective as bringing 20% more hit points per ship. 

This makes sense intuitively because each additional pilot that you bring is also bringing more damage capability as well as more hit points that opponents have to chew through to remove that damage capability from the field.  Everyone who is experienced with EVE combat knows that it is good to bring more pilots if you can, but we now have a quantitative way (albeit under nominal conditions) to compare the trade-offs between the quality of pilots/ships, and the quantity of pilots/ships.

Let's call this number the Fleet Figure of Merit,
$\Large FOM = N \times \sqrt{DPS \times HP}$  Eq. 2

So, you get more mileage toward victory from ship quantity than you do from ship quality. I don't need to provide examples of this to any reader familiar with the history of conflict in EVE, although being able to show it with clarity is new to the EVE discussion. Fleet commanders who are risk averse stubbornly avoid fights where they are outnumbered precisely because number imbalance results in rapid losses, as this simple dynamic model confirms formally.
Returning to the dynamics equation Eq. 2–4, which I wrote in terms of the numbers of ships in both fleets, I can now write this with more insight into the factors that determines the outcome of the fight. Rewriting the dynamics equations in terms of the Fleet Figures of Merit and how these change with time by simply rearranging the constants,
$\large N_a(t) \sqrt{DPS_A HP_A} = FOM_A cosh(t/T_{mutual}) - FOM_B sinh(t/T_{mutual})$
$\large N_b(t) \sqrt{DPS_B HP_B} = FOM_B cosh(t / T_{mutual}) - FOM_A sinh(t / T_{mutual})$  Eq. 2–7

This viewpoint suggests that we can think of an engagement as a competition between two dynamically updating figures of merit, which in an aggregate statistical sense, is a useful guide for thinking about large engagements.  

Discussion and a word of caution

In the derivation above, I used two state variables, the number of ships in each fleet, and two sets of constants, damage per second and hit points, to model dynamics of two fleets in combat.  This model for the dynamics of fleet combat assumes that everyone is doing their job and that fleet compositions are homogeneous.  The model suggests that engagement  outcomes under these circumstances are statistically determined by a figure of merit that has squared benefit for pilot numbers in comparison with ship quality or pilot skill points.  

I have made no effort to show that this model, based on assuming all pilots take commands in the same way, will be equal to the experimental average of fleet combat in game.  So this is a good place to caution the reader on the limitations of seeing fleets as only Figures of Merit.  As with all of the conclusions based on abstract models, the effective damage rate, and even the effective hit points depend on what you are fighting because of target resists, range, speed and signature radii.  As I will illustrate in an example below and in a following post, the $FOM$ number hides many limitations. Successful fleet commanders have to balance the qualities of the ships that are reflected in this number with those properties that are not, such as mobility, damage projection, tackle and electronic warfare. Using $FOM$ to analyze, say, Alliance Tournament teams, might be misguided.

Next, I want to elaborate on a metric that fleet commanders can use (and already do use) during the fight to assess the status of the engagement.  

Clues in the Rates

The challenge with applying the work I have presented above is that the effective damage rates and hit points are almost impossible to judge prior to a fight. Even if you had perfect information about the numbers and fittings of a hostile force, their skill at focusing fire, applying logistics and electronic warfare, would be unknown. Instead of trying to use only scouting information, how can they help us during a fight?

Gathering information takes time, something which fleet commanders have little of. Is there an easy way to determine the fleet advantage during combat? If it is too time consuming for a fleet commander to determine this information without assistance from fleet mates, what is the right digest of information that he or she can request from lieutenants to help make decisions about whether to fight or GTFO (i.e. get the fleet out)? 

Some fleet commanders will ask that pilots of a certain ship type put an 'x' in fleet chat after they have lost their ship so that everyone can see how quickly the fleet is losing ships. If you watched the Zarvox T1 fleet that I linked earlier in the article, you probably also noticed that he requests, with repeated urgency, that pilots who lost their ships report it in fleet chat.  This method is trying to gauge the situation by looking at a rate of ship loss.

Other fleet commanders will ask that one pilot report to them at the time that they have lost a certain number of ships of a certain type.  In the 'Engaging & Disengaging' edition of Jin'Taan's 'Fundamentals of FCing' series, he discusses several factors related to when to decide to leave an engagement.  Among many other considerations in the 'Situational Awareness' section, he uses a comparative loss rate between fleets to make that decision.

The insight from these very experienced FCs is consistent with my simple dynamical model for homogeneous fleet engagement.  I claim that their intuitive approach of comparing rates of loss is, in fact, the same as comparing the dynamically changing figures of merit. That is, comparing the fraction of ships lost in a given time by each fleet is the same as comparing the effective fleet figures of merit.  This is true for homogeneous fleets of equal or unequal size.  

Intuitively, you are computing $\Delta N_a / N_A$ and comparing it with $\Delta N_b/N_B$ for a given unit of time.  If Fleet A has 30 ships, and Fleet B has 10 ships, the loss rate of ships in A must be less than three times those of Fleet B for Fleet B to hold the field. As with most of the conclusions in this note, it should come as no surprise that the emergent behavior of successful fleet commanders is already achieving the same objective function for time analysis of combat dynamics as I derive from the dynamic model.

Looking at the proof of this approach, by writing the rates from our Eq.2-4, at a time $T_s$,
$\large \left ( \frac{1}{N_a(T_s)} \right ) \left. \frac{dN_a(t)}{dt} \right |_{T_s} = \frac{DPS_B N_b(T_s)}{HP_A N_a(T_s)}$
$\large \left (\frac{1}{N_b(T_s)} \right ) \left. \frac{dN_b(t)}{dt} \right |_{T_s} = \frac{DPS_A N_a(T_s)}{HP_B N_b(T_s)}$

When we compare these as a ratio, the result is the ratio of the fleet figures of merit sampled at time $T_s$, or $\frac{N_a^2(T_s) HP_A DPS_A}{N_b^2(T_s) HP_B DPS_B}$.  This is a dynamic update of how the actual damage rate and number of ships are having as an effect on the potential outcome.  What adds to the challenge of keeping running totals of the rates of ship-loss in your fleet and hostile fleets is that in most complex engagements, hostiles can enter and leave the field constantly. Fleet commanders update their internal models for the status of hostile and friendly forces based on the rates of ship destruction.  Specialized overview setups can help with small engagements, however, large engagements would need a more dedicated approach, which is usually not practical. Perhaps some better tools could be written along these lines? 

When you have a logistics wing in your fleet, you have the advantage that a dedicated group of pilots are constantly examining survival timing for ships in your own fleet. Some fleet commanders try to gather information from this group by asking whether, "reps are holding?" Some fleet commanders rely on this type of information as well to get a sense of the dynamics of the engagement. 

An Applied Example

I want to share an example of how features of these dynamics manifest in real fleet combat situations. This is the most dangerous part of this blog post because readers may misinterpret my intent here.  My objective in sharing this example is just to show that; (1) There can be qualitative (and in some cases quantitative) agreement between the simple dynamical model, and; (2) That rates of ship loss are diagnostic of the fleet figures of merit.  

I have chosen to consider a battle between Imperium and FAS dreadnoughts in July 2017.  Choosing a capital fight might have simplified the discussion in this section because siege-mode combat means that remote logistics will not interfere in effective HP assumption and it also means that ease of tackle means that we can watch the fight until the bitter end with one side being completely destroyed.  

The fight in question is occured in MDD-79 on July 11th, 2017 from 18:15 to 18:29.  There is also a video record of this engagement.  There is also a subcapital fight happening at the same time, and as you will see in the analysis there is cross-over between these that affects the outcome, but I will focus on the dreadnought numbers only.  The damage dealers present on grid for both fleets at the start of the fight are:

Fleet AFleet B
Mixed Dreadnoughts1930

The mixture of capital types may cause readers to question the assumption that both fleets are homogeneous, however, I argue that dreadnoughts are intentionally chosen to have relatively similar hit-point and and damage rates.  They are similar enough that it should not affect our conclusions.  From the battle report data, I determine the total remaining ships at 60 second intervals and plot these in Figure 5.  These curves exhibit qualitative characteristics of the mutual-predation model, which only goes to show that our assumption about constant loss rates is true within a margin of error.  I would restate this as large fleets behave statistically.  If we look even deeper, there are clues about the performance of both fleets.  

Figure 5: Number of dreadnoughts surviving for both fleets is shown versus time.  Also shown is the model dynamics based on average parameters for both fleets determined from kill mail records.

For starters, lets plot the results of what the dynamic model shows as the total number of remaining ships as time moves forward.  These dashed lines show that Fleet B (Imperium) lost rather more ships than would have been expected to accomplish complete elimination of the FAS dreadnoughts.  

Looking at who was shooting who during this period resolves this question.  Reviewing the number of involved ships on all kill mails reveals that the FAS battleship fleet was also targeting dreadnoughts during the 150s to 400s period, trimming Imperium numbers.  Such a choice of course comes with other trade-offs considering there was also an Imperium battleship fleet present on grid as well.  

This model-data playback does not achieve quantitative agreement and this is because this fight violates some of the initial assumptions.  From the outset, the mutual-predation model will be helpful when all damage sources are accounted for.  This example does serve to show that fleet numbers behave statistically when fleets are even moderate in size.  Second, it shows that when the assumptions of homogeneous fleet combat are met that mutual-predation dynamics dominate the time-evolution of ship losses and overall combat outcome.  Stated in terms of the model parameters, numbers of ships, hit points and damage per second, the rates of ship loss correspond to the ratios of fleet figures of merit. 

The outcome of this engagement was largely inevitable given the numbers and timing, and I chose this example for simplicity.   Thus the model does not suggest any change in tactics for either fleet (apart from not getting dropped by a large number of Imperium dreadnoughts), a dynamical model should enable us to identify possible choices where alternative uses of auxilliary or subcapital fleets might make other outcomes and opportunities possible.  This is a much larger topic, and beyond the scope of this note. 

Now that we have seen an example, what other uses can we imagine for a model for homogeneous fleet engagements?

Fleet Capability: Can We Quantify Risk Taking?

Can we use these results to determine the number and composition of enemy fleets that I can engage safely?  Let's imagine that I am in command of a fleet and I am willing to lose some fraction, $S$, of my fleet.  What are the properties of hostile fleets that I can safely engage without exceeding this loss ratio?  

Before you discard this questioning as being inconsistent with how you command fleets, consider that if you know how your opponent is answering this question, you might use the result to determine what fraction of your numbers to hide to entice a fight that you can win!  Of course, don't forget the amount of time it takes for your reinforcements to come to your aid. 

If I calculate the number of Fleet A's ships alive at the time when the Fleet B's numbers reach zero, at $t_{death,B}$, the fraction of ships surviving is,
$\large S = \frac{N_a(t_{death,B})}{N_A} = \sqrt{ 1 - \frac{N_B^2 DPS_B HP_B}{N_A^2 DPS_A HP_A} }$

This is valid for $FOM_A > FOM_B$ but the goal here is to make Fleet B think that $FOM_B \gt FOM_A$.  If you believe that a hostile fleet $B$ with $FOM_B$ will accept a loss of $S$ fraction of their fleet, you would choose to hide a fraction of your numbers to achieve,
$\large FOM_A \leq \frac{FOM_B}{1 - S^2}$

So, in this context we can think of $S$ as the risk aversion of a hostile fleet commander.  Ranking fleet commanders in this way could produce an interesting survey across New Eden.  Given the information from scouts, what target FOM are they willing to engage compared to their own fleet?  Of course, most of the time this will be overshadowed by logistics and electronic warfare considerations, so a practical survey is not likely using this simple model but it is interesting to consider. 

Executive Summary

  • In EVE, as in life, time is the only thing you can't go get more of.  When fighting opponents in same-size or smaller-size targets, therefore, prelocking secondaries can have a dramatic effect on combat effectiveness.
  • For homogeneous fleets in combat, the rates of target destruction are determined by the ratio of each fleet's figures of merit, or FOM: $FOM = N \times \sqrt{ DPS \times EHP}$.  Some notes on the FOM concept:
    • The number of ships in the fleet has squared benefit over ship quality, i.e. $DPS \times HP$ product.  This makes obvious intuitive sense considering that bringing one more ship means you have also brought more $EHP$ and more $DPS$, both. 
    • Ship quality factors, i.e. effective hit points and effective damage per second, are interchangeable in this dynamic model.  In reality, of course, they are not because our definition for $T_{kill}$ in this model assumes that the logistics repair rate is constant. 
    • It is usually impractical to calculate a fleet figure of merit during combat.  Instead, by comparing the fractional rate of loss for your own fleet and your opponent's fleet for any time period, you are comparing the effective fleet figures of merit. Methods along these lines are already widely used. 
  • With a model for fleet advantage, it is possible to quantify the risk aversion of fleet commanders by comparing their fleet FOM against hostile fleets FOM where they choose whether or not to engage. 


I want to thank Jin'Taan and Elo Knight for assisting me as a sounding board for ideas over the past months and years.

Appendix A: The Dumbo Factor

Can we account for the fact that not all pilots in damage-dealing ships will be doing what they are told?  Some of them may be eating a burrito.  This can affect how the lock time for the target, as well as the fraction of pilots applying damage to the intended target.  Famed fleet commander and fluent francophone, Elo Knight, refers to this as the "Dumbo Factor".  

We know that our pilots aren't perfect, so we can try to lump their imperfections into some of the functions we have used in the derivation above.  If we were to encode this inefficiency in our DPS model for the $T_{kill}$ expression, it would include both a locking-time delay factor, $T_{dumbo}$, and account for a number of pilots in the fleet not applying any DPS to the target, $N_{dumbo}$.  Our new expression would be, 
$\large T_{kill} = T_{lock} + T_{dumbo} + T_{flight time} + \frac{HP_{target}}{ \left ( 1 -  \frac{N_{dumbo}}{N} \right ) DPS - RPS}$ 
In this way, we can factor pilot imperfection into our effective DPS consideration.  In Figure 1, I illustrate a hypothetical time-evolution of the damage for a target-calling event, assuming the target is broadcast at $t = 0$.  We can easily group ineffective pilots into the effective DPS and locking time of the fleet.   Of course, the dynamical model presented in this note does not include finite delay in locking, so some extensions may be needed going forward.  
Figure A-1: Example DPS plot versus time with one-second intervals.  In this example, I show $N=100$ ships each doing one unit of DPS, and $T_{lock}=4s$.  I have also illustrated how the 'Dumbo Factor' could influence this profile by including $N_{dumbo}=5$, and $T_{dumbo} = 2s$. A Poisson arrival process approximates the lock-time of imperfect pilots although a model with added delay of $T_{dumbo}$ would likely be sufficient to capture the performance of the fleet.

Thank you for reading. 

Sunday, April 16, 2017

Frigates of EVE: A New Physics Model

 Like many of you, I will be ordering the "Frigates of EVE" book from the EVE-Online store. Max's talk about the PEG at FanFest is here.  As many of you know from communicating with me in person, I have not participated in lore discussions, as I only work with models that have predictive value -- having a story about how it fits into the history of the game has been a secondary consideration.  Having said that, I had a few brief thoughts on this, since it involves the physics of EVE and I am notorious for working and writing on this subject. 

 First, based only on the presentation at FanFest we have learned that anything that warps, cloaks or is anchored in New Eden is powered by one or more power enhanced generators (PEG) which each create a 'subspace singularity'.  The details may be more carefully revealed in the book, but I infer that acceleration or anchoring effects arise from either a relativistic mass effect or from an interference pattern that can be tuned to either constructive or destructive interference.  Of course, I will wait to read the chapter itself before deciding if I add any value to the communication on this model.  

 Second, the model does not alter the first-order differential equations that govern ship motion which I presented in Chapter I and in all my subsequent blog postsThe justification for the 'inertia' coefficient now is explained by the presence of a subspace singularity.  Thus, drag is produced on any EVE ship that can warp and/or cloakNone of the applied physics work that I have presented will need to be fundamentally revised, although changes to vocabulary would be helpful for lore enthusiasts.  We can still use all of the tools that I have presented and developed to analyze subwarp ship motion and ship-to-ship interactions like orbits and bumping.  

 Third, the presentation of the PEG model is encouraging because it shows that CCP recognizes the importance of the physics model that underlies ship motion in EVE online.  They have invested in explaining the root cause for its most distinguishing phenomena and are committed to itWhat attracts me to the classical explanations of EVE's model is that it lends itself well to predictive models, player intuition and the many tools that have been developed for linear systems analysis.  If this PEG model makes EVE's linear drag mechanics palatable to the lore community by explaining why there is maximum velocity for certain objects in space, then I am enthusiastic about receiving it.  

 Finally, the model presented at FanFest 2017 answers some questions about how the New Eden universe works but it also raises questions and opportunities for future additions to the game.  One of the obvious opportunities that has presented itself is to imagine ships that do not include any PEGs and therefore do not produce a subspace field.  Such ships would be limited only by the speed of light, or whatever other governing particle/energy is constituent to New Eden.  I do not yet know if this is discussed in the Frigates of EVE physics chapter, but I certainly hope we consider this opportunity.  I discuss some of the properties of such ships belowI look forward to this addition to EVE.  

PEG-Free Ships
 Imagine the interesting role such ships would play.  While these ships would not be able to warp or cloak, they could be launched from stationary structures, citadels or even large ships.  They would have finite capacitor energy but they would be able to accelerate continuouslyIf we assume relativistic effects for the motion of such a craft, their maximum velocity would be limited only by the speed of light, and the amount of energy that you could carry in the form of fuel.  You would not be able to pilot them with a pod because pods include PEG devices.  A cockpit would certainly be possible, or they could be piloted by a drone computer of some sort.  It seems that this is likely the motion model for the Valkyrie ships?  During the Paschal holiday, I took a few moments to think about the energy bounds of these hypothetical PEG-free ships.

Without an infinite source of energy, we have to imagine that such ships would have to carry their energy with them.  They cannot rely upon the infinite PEG energy that slowly recharges a capacitor.  Lets also suppose that we wanted them to carry armaments comparable to a frigate, requiring it to be a similar mass to a frigate and similar energy requirements.  With finite fuel limitations, EVE pilots would want to use such ships for tackling ships near stations and other deployment locations.  So, the capability to use warp scrambling and webbing technologies would be desirable.  

The most energy dense fuel I can think of today is antimatter.  I wont go into ideas on how you could make use of antimatter for propulsion and self-containment at this time, but I am confident that 23,000 years from now these issues would be have been well addressed.  So, we have a ship with a mass of 106 kg, that needs the energy to accelerate continuously for, lets say, an hour of flight time before docking to refuel.  

Now, I'm not advocating that PEG-free ships be able to approach the speed of lightAs readers are no doubt aware, reaching velocities approaching the speed of light begins to require an exponential more energy (see Figure 04-17-a).  I have highlighted the 6000m/s datapoint on the figure to show that there is significant region to improve EVE ship velocity without requiring ship mass to be overwhelmed with antimatter fuel mass requirements. 

Figure 04-17-a: Energy needed to accelerate an object with mass 106kg to a velocity, v.  Note that for velocity to reach the speed of light, you would need infinite energy, which is not shown.

Figure 4-17-b: Mass of fuel needed to accelerate a mass of 106kg to relativistic velocities is shown.  Achieving 10% to 20% of the velocity of light could be done without impacting overall ship mass. 

Figure 4-17-b above shows that we can achieve millions of meters per second in velocity powering our frigate propulsion with antimatter annihilation instead of with PEGs. I am willing to concede that antimatter containment and propulsion are constrained by thermodynamic efficiency limits even 23,000 years from now, but even with less than 50% efficiency the total fuel mass is plausible for a PEG-free frigate. As an aside, if such a frigate were limited to the fuel energy density of today's technology, i.e. liquid hydrogen with 142MJ/kg, the mass requirements for relativistic velocity would be staggering and unworkable.

So if we aren't practically limited by maximum velocity, might we be limited by maximum acceleration? Lets assume the rate of acceleration is limited to the highest acceleration of the most agile ship in EVE, which recently was a microwarp drive fitted Dramiel capable of roughly 100G acceleration. Of course, depending on how long your memory is, you know that ships in EVE actually have had much higher accelerations in the distant past.  In the age of speed-fits, prior to the Dominion rebalance, dedicated interceptor fits could achieve 800G of acceleration, with a top speed of 24,000m/s. So, we know that velocities in that range are workable in the EVE engine. 

In my October 2015 poast, I updated the notes on fundamental limitations of orbits. In that note, I showed that close in orbits are limited by acceleration, not velocity, so deploying a PEG-free frigate for high angular velocity tackling would be far superior to PEG ships. For on-grid tackling near stations or capitals, PEG-free ships would be unmatched for tackling or skirmishing. Sustained acceleration of 800G would allow a orbit angular velocity of 3 radians per second, an order of magnitude higher than we can achieve with existing PEG ships.

From a lore perspective, you might imagine that the fluid in a pod could protect the pilot from acceleration to some degree. Indeed, if the fluid inside a PEG-fitted pod can protect a pilot from this sort of acceleration, we can imagine a PEG-free pod that would provide similar protection. Experiments on humans did not reveal a specific short-term acceleration limit for the human body. I suspect that the many years of technological development and modifications of the body that are inherent in implants would have vastly extended the capability of a pilot's body.

In summary, now that we know how PEG ships work, I look forward to the inclusion of PEG-free ships to EVE. I for one will be purchasing the Frigates of EVE book, and reviewing the motion model chapter with keen interest in places where we can make predictive improvements in how ships behave in strategic situations. If the physics chapter includes a proposal for PEG-free ships, then I will consider this model complete.

I hope you have enjoyed these quick thoughts about Frigates of EVE and the physics model presented at FanFest 2017. My next post will be the first of a two-part offering that explores a very different topic in EVE, one that is not related directly to motion or applied physics but instead asks how we make decisions in combat. Thank you for your readership and for comments on this work.

S. Santorine


(Just another link,
 to make you think,
 about exhaust versus thrust.