Preliminary Notes on Colonizing the Universe

Authors note: Roughly correct as far as I know but I’m not a physicist so there certainly might be mistakes or things that I have missed. Obviously extremely speculative about far-future scenarios, but I think this provides an interesting high level ‘timeline of the far future’ to start thinking through. This all assumes current physics/cosmology, especially today’s standard slowly-expanding LCDM cosmology. If our cosmological or physics understanding changes this can shift the scenario dramatically.

Recently, I was waiting for the train one morning and started pondering the actual mechanics of colonising the universe. Obviously these thoughts are not new and e.g. Nick Bostrom, Stuart Armstrong, and Anders Sandberg et al have done an awful lot of work here on the basic arguments showing that intergalactic colonisation with von-Neumann probes is feasible and won’t take even that much energy. However beyond just sending out the probes there is a whole question of what do the probes do once there, how can they send mass/energy back to us, and what a mature K4¹ civilization would look like in practice. My goal here is ultimately to try to synthesise a bunch of different ideas into a unified timeline and operational plan for the far future colonisation wave and to work out at least some extremely speculative engineering of how we would go about it in practice.

Let’s begin with some basics. Current physics / cosmology posits we live in an expanding universe with an accelerating expansion rate. This, combined with a fixed speed of light, means that there is a fixed causal volume that we can ever interact with. Outside of this volume the galaxies are either receding from us faster than light or will be by the time anything we send to them could reach them. Thus they are irrevocably causally disconnected from us. We can no longer interact with them in the future, nor they with us, although light they emitted earlier in time can still reach us, so we can see them, and vice versa.

This means that if we plan to colonise the universe and ultimately do something meaningful with its free energy, then there is a finite causal volume over which we can do this. Moreover, this volume is shrinking every year, as the expansion of the universe continues. This is the astronomical waste argument. This rate of loss is not that dramatic, every year we lose perhaps O(1) galaxy which is obviously a massive loss in absolute terms, but not relative to the tens or hundreds of billions of galaxies within our existing causal volume. However, if we wait billions of years to do anything our accessible volume shrinks dramatically. If our ability to instantiate our values is proportional to the free energy we have accessible, then by letting galaxies escape over the causal horizon, we are forfeiting immense – ‘astronomical’ – amounts of potential future value.

This naturally leads to the idea that we need to send probes out to colonise these distant galaxies. Sending out ‘normal’ spacecraft is obviously too slow so we have to send out tiny von-Neumann probes which travel at a high fraction of the speed of light. These are effectively ‘seeds’ of our entire civilisation where, once they reach a target galaxy, they rapidly self-replicate and begin to manufacture from scratch effectively their own K3 civilisation with purely local resources.

As a rough estimate, there are of order 1 - 10 billion galaxies within our causally accessible region of the universe, and somewhat more if we count dwarf galaxies and other nontrivial matter concentrations. This means we likely have to launch O(10^9-10^12) probes, depending on our assessment of probe survivability and level of conservatism about missing galaxies, during the length of our colonization program. This sounds like a lot but even still it pales compared to the resources of even a basic K2 civilization.

Basic probe design

Okay so firstly let’s assess how feasible this is; what are the various important parameters, and what are the engineering challenges?

The first thing to note is that the speed at which we dispatch the probes is extremely important. This is because the universe is expanding while the probes are traveling, so that if the probes travel slower, the longer travel times means they cannot reach distant destinations before they fall over the horizon². Because we live in a 3 dimensional universe, the accessible volume scales as the cube of the radius, where the length of the radius is effectively the speed that we can travel at. This cubic relationship becomes incredibly important because it means that lower speeds are much worse than you would naively think. For instance, if our probe travels at 0.1c then it can reach only 0.1^3th, or 1/1000th the total causal radius that a probe travelling at ~1c could. This is terrible. At 0.2c we can only reach 0.8% of the total theoretically accessible volume. At higher fractions of c, however, this ratio reverses. By 0.9c we can reach 72% of the theoretical maximum and at 0.99c we can reach 97%. Beyond that there are obviously steeply diminishing returns since we must asymptote at 100% of the volume. Moreover, higher fractions of c require dramatically increasing energy costs, shielding and deceleration challenges and other issues, so there is a trade-off here. Naively, I would think that somewhere in the 0.9-0.99c is optimal if the other engineering challenges can be addressed.

For now we are going to assume a roughly 0.9c terminal velocity for the probes. What do these probes look like? Firstly, let’s recap what the probes have to do:

1.) Be able to be accelerated at a pretty high clip to relativistic velocities in the home system.

2.) Survive for billions of years cruising through deep space under potentially all kinds of conditions.

3.) Survive cruising at relativistic velocities where even tiny collisions become extremely energetic and all radiation facing the probe is heavily blue shifted and thus is more energetic and damaging.

4.) Navigate to the target galaxy, decelerate from relativistic velocities to effectively stationary In the frame of the target galaxy, and then navigate to land on a suitable target in the target galaxy.

5.) From whatever planet/asteroid etc you landed on, somehow bootstrap a K3 civilization quickly from scratch.

6.) Maintain enough stored information and alignment that they follow the original instructions from home vs diverging and doing something weird.

Here I’m going to mostly focus on 1-4 and ignore 5-6. This is because points 1-4 contain most of the hard physics constraints and can be most understood today while 5 and 6 are harder computational problems of what information and knowledge you need to store to keep the probe aligned and let it know how to build a K3 from scratch. All I’m going to assume is that this ‘seed’ is fairly compact — i.e. a few tonnes and can handle decent but not insane accelerations. So 1.) is primarily a constraint on the launcher. We can achieve any desired acceleration by making the launcher longer, but at some point this just becomes ridiculous. An ambitious K2 can likely build a launcher hundreds of AUs long which corresponds to only hundreds of gs acceleration along its length to get to 0.9c. I’m just going to assume that this suffices. Requirements 2.) and 3.) are both very challenging, but in different ways. Starting with 3.) this essentially means that the ‘forward facing’ direction of the probe will have to be extremely well shielded. This will likely include the vast majority of the mass of the probe being extremely hardened sacrificial shielding. There are essentially two primary threats during an intergalactic cruise. Firstly, a constant bombardment of high energy photons of ambient starlight which has been blue shifted, and secondly rare but potentially highly energetic collisions with dust grains and micrometeorites (I assume that the probe has enough sensing and navigation capabilities to avoid hitting anything with actual substantive mass such as a dense hydrogen cloud, or an asteroid etc, since such impacts would likely be fatal).

Most likely we will want several layers of shielding, likely resembling Whipple Shields arranged with high redundancy. We can envisage the probe looking kind of like a very long thin tube, or needle, with almost all the forward mass being shielding of some kind. Having as small a cross-sectional area facing forward is paramount, since this is ultimately the primary surface that comes into contact with the relativistic particles and dust-grains. Likely we would want <1m front-facing area if possible. In addition to kinetic shielding, we would likely want magnetic shielding in addition to deflect charged particles ahead of the path of the probe where possible. However, magnetic shielding does nothing to neutral particles, such as hydrogen atoms, so we have to decide whether to tank these via ablative shielding or use some kind of active ionisation method to ionise them into charged particles then deflect. Likely, we will want to do both to have multiple layers of redundancy.

Specifically, this looks something like:

1.) Forward facing lidar/radar sensing systems to map oncoming mass distribution with the goal of detecting and steering around any large concentrations of mass (e.g. a gram sized clump would be a catastrophic impact and so must be avoided), and detecting smaller mass sizes for ionisation

2.) Ionisation system, probably laser based, to vaporise/ionize any smaller mass clumps. The goal here is to make them charged so they can be deflected by the magnetic deflection system.

3.) Magnetic deflection shielding. This will deflect any ambient charged protons/electrons in the cosmic background, plus any neutral dust that has been ionised.

4.) Physical ablative shielding covering making up most of the forward length of the ‘needle’. This is designed to absorb impacts by relativistic neutral particles such as photons or neutrons which cannot be deflected, and secondly to tank impacts by any other particles which fail to be deflected for whatever reason. Secondly, it is designed to be the passive shielding of last resort in case the active shielding methods such as the magnetic deflection and ionising-laser systems fail, which there is a decent probability they will eventually.

For a 1 tonne payload mass, we probably need 10-50 tonnes of shielding and secondary systems mass to be somewhat safe. Moreover, we will need to be very careful with the flight path to not pass through dense regions such as galaxies, galactic halos, inter-galactic dust clouds etc but stick to the extremely empty intergalactic void medium for as much of the flight as possible. Surprisingly, if primarily travelling through an extremely sparse intergalactic void, heat dissipation and drag are not that bad, although a serious design must obviously include allowances for these.

Coming back to point 2.). Notice that this probe is not a dumb passively flying slug but has on-board all kinds of active sensing and deflecting systems, as well as the core computational unit of the payload itself which has to maintain sufficient capability to land in a distant galaxy and start up a K3 civilization from scratch there. This means we need to have high-precision electronics surviving for billions of years in a fairly hostile (even with the shielding) radiation environment.

This is then the cost of active shielding and sensing. Active shielding could substantially reduce the necessary mass of passive shielding, and increase survival rate, but introduces much higher probability of failure of the active systems. A massive mass shield cannot ‘fail’ in the same sense but is much more expensive to launch, and is helpless before random events e.g. it cannot notice or steer away from an unlucky micrometeorite in its path.

Designing complex electronic and mechanical systems which can operate for billions of years without failure is obviously extremely challenging. I cannot say it is impossible for a K2, but most likely, the probe would be designed with many redundant subsystems so a single failure does not paralyse the entire probe. Secondly, for core civilisational information, it must likely be inscribed with high redundancy on both active error-correcting systems as well as physically on extremely hardened materials. Moreover, the probe must likely need ‘living’ and self-healing properties and be capable of manufacturing spare-parts and new instruments as necessary. I.e. It does not necessarily carry a fixed contingent of parts but rather it possesses a large reservoir of feedstock mass and then advanced manufacturing capacity to recreate damaged components and perform repairs in-flight, rather than relying simply on pure redundancy. Obviously, the manufacturing equipment itself could fail and this possibility will need to be handled with additional redundancy also. This is not as bad as it seems however, since we assume that the probe payload must already have extremely advanced manufacturing capabilities since it is expected to land and be able to begin creating a K3 civilization at the destination. Likely, these manufacturing capabilities can be used internally to maintain probe health during its interstellar flight too. This is all far beyond the engineering we are capable of today, however we have to assume there are no fundamental show-stoppers.

There is also the power question. Not all components can be purely passive, although all electronics etc should be designed for low-power, relatively slow action. This means the probe will need some kind of active power source that lasts for billions of years. Luckily, this is not that fundamentally problematic. Most likely, for relatively small power draws, the probe can work with an extremely-slow-burn radioisotope decay reactor. There are isotopes such as U-235 and U-238 with half-lives of hundreds of millions to billions of years which can be used to power the probe throughout its expected lifetime. However, due to the extremely low half-life of these isotopes, the power density is also extremely low. This means that the probe would either have to contain a very large amount of fissile material (increasing mass budget), run on very low continuous power draws (in the milli-watt to watt range) or need an active reactor vs passive radioisotope decay which can increase power-density at the cost of yet more active equipment which needs redundancy, and can fail. It is also possible to combine these systems with a slow-burn purely passive radioisotope decay power system providing milli-watt power continuously throughout the cruise with a mostly-dormant fission reactor which can provide ~1-10 watts of power when active which can be turned on when extra power is needed for e.g. specific self-repair jobs or when entering a more active navigation phase or when decelerating around the target galaxy.

One minor note: the probes themselves need to have very sophisticated navigation and steering capabilities. This is because during launch, especially to distant galaxies, we only have extremely delayed information to do the targeting on. If we are targeting a galaxy 10 billion light years away, our only knowledge of that galaxy is from 10 billion light years ago, at which point we might only see the very earliest stages of galaxy formation. Moreover, by the time the probe reaches the galaxy it will be 11 billion light years later, so if we launch blindly, we are having to make a prediction accurately over 21 billion years (!). Of course, we would expect the K2 civilization launching the probes to have very good cosmological models to predict where the galaxies it sees now will eventually end up, but there are fundamental limits to prediction especially over these extremely long timescales.

This obviously doesn’t mean that sending out probes is doomed, but rather the probes themselves cannot be dumb slugs that we just shoot at a particular object in a far-distant galaxy. Instead, we can realistically only shoot them at large windows of ‘predicted future mass density’ and then they have to take over navigation from there. This is not as demanding as it sounds. As the probe approaches the target, assuming they have any decent sensors, they will get increasingly up-to-date information as they approach. Moreover, even at 0.9c, the lateral velocity needed to change angle of arrival if you start making course corrections billions of years away is fairly modest. Nevertheless, we cannot think of these probes as some kind of dumb inert nanobots. Rather they will need to have sophisticated sensing apparatuses and steering capabilities, potentially onboard rockets or other means of propulsion. This means the probes must be heavier than you would think and can likely also withstand less acceleration than is possible with a fairly compact slug. This is not fatal, it just requires a longer launcher in the home system.

Even with all this shielding and redundancy, it is likely that survivability will still be pretty low. The shielding, sensing, and then final deceleration are the primary engineering challenges and failure-points in the probe’s mission. There are just too many things that could go wrong — components failures over billions of years, immediate death by hitting something micrometeorite sized at relativistic velocities vs just intergalactic dust, failures of sensing or navigation, etc. This means that we probably have to launch 100 or 1000 probes per galaxy we target to be safe. This obviously increases the amount of probes we have to launch from the tens of billions to the trillions, but even still this is well within the means of even our K2 civilization. It is also worth pointing out that theoretically we only need to have one probe survive per local group. Once a probe has arrived in some local cluster, it can send out its own probes to replicate in other galaxies within that cluster, as long as the cluster stays gravitationally bound. This will slow down the general colonization wave slightly, and thus cost some mass/energy, but is a good backup plan, because certainly even with heavy redundancy some galaxies will end up missed by the initial launches.

Launching

Now we have discussed the probes, we need to do some exploratory engineering on the launchers. The requirements for these are pretty simple. We need a system that can:

1.) Launch a ~50 tonne probe to 0.9c.

2.) Launch with sufficiently low acceleration not to damage the internal mechanisms of the probe.

3.) Ultimately launch trillions of probes over thousands of years.

Firstly, the raw energy and mass requirements are not that bad. Mass is pretty trivial. Even a trillion 50 tonne probes is only the size of a large asteroid in mass. Obviously, the probes will not be pure random rock but extremely carefully machined objects. There should still be plenty of raw materials in the solar system to cover this though. Energy costs, though large, are also not prohibitive. Assuming we are a K2 civilization, even trillions of launches only requires approximately 0.1% of Sol’s energy over a few thousand years. This is nontrivial but pretty cheap. We are certainly not bottlenecked on colonising the universe by energy.

To actually launch, however, we need to accelerate a single object to 0.9-0.99c, which is a highly nontrivial amount of energy.

There are basically two separate but potentially complementary launch methods: magnetic linear accelerators and laser sails.

A magnetic linear accelerator is similar to a particle accelerator or a railgun here on earth. The goal here is to shoot the probe through a series of extremely strong magnetic fields which accelerate it, without touching, since nothing would survive contact at these relativistic velocities. The linear accelerator is extremely good at imparting immediate acceleration into the probe at high efficiency. Moreover, an accelerator can launch a large amount of probes in quick succession by pipelining them along its lengths. However, for reasonable accelerations, the accelerator has to be extremely long. Certainly in the AU scale and potentially into the hundreds of AUs scale — so this in practice could be a chain of stations from Dyson sphere to the Oort Cloud each precisely phased to activate as the relativistic probe passes. This is achievable for a K2 civilization but is nevertheless a substantial infrastructure project³.

A second approach is to use beamed power directly on a solar sail around the probe. This does not require fixed infrastructure, although it does require both a very powerful beam and a clear path for the beam to operate in. The amount of power needed to accelerate a probe to relativistic velocities is extremely large and thus the beam becomes a massive safety hazard given there is likely a large amount of other K2 infrastructure in Sol. The beam potentially also allows high efficiency conversion of photon momentum into velocity for the probe, although its efficiency falls dramatically at high probe speeds because the photons arrive already substantially red-shifted. It requires the probe to have a large solar sail which would likely be discarded once the beaming is stopped, likely as the probe exits the galactic halo. A massive advantage though is that since the beam does not require fixed infrastructure, the acceleration can realistically go on for a very long time across a very large distance. This means that it can target much lower accelerations than the linear accelerator can realistically achieve. This is important if the probe contains extremely delicate machinery or electronics which cannot handle high g-forces.

It is also possible to have hybrids of the launcher design. For instance, a linear accelerator from Sol and then beaming power-stations out in the galactic halo to provide velocity boosts or top ups, additional steering power, etc. This may be important for regions of sky that are hard to directly reach from Sol.

Launch location is also important. Theoretically, if we want to start launching we can do so immediately from our home system. The advantages of this are that they are where our core infrastructure is likely to be, power is easy since we have the sun to dysonize, and we don’t have to travel. The downsides are primarily that we are already in the middle of a galaxy. This means that the probes we send out have to first exit our galaxy before entering the intergalactic void. This will cause inevitable, and potentially large, ‘infant mortality’ for our probes since inside the milky-way there is a substantially higher mass concentration than outside of it. Secondly, although we can reach most of the sky from sol, there are some directions which involve shooting through the plane of our galaxy which we cannot realistically cover.

There are two options here. Firstly, we can launch directly from Sol and just deal with the infant mortality. To make things easier in our surrounding galaxy we can also send out ships or just massive lasers ahead of the probe to try to clear a path through the dust and reduce its density along designated launch lanes. To completely solve the problem we could also just construct launch stations above and below the galactic halo and launch from there. By the time we have constructed these stations though we would likely be approaching K3 rather than K2 status, so probably the initial wave should depart from Sol and then if we need to send more waves as we expand we can construct the full extra-galactic launchers.

A lot also changes depending on the distance of the target galaxy. Over long distances the probe will be slowed down not by drag but by the overall expansion of the universe. The probe likely does not have the energy to maintain acceleration against the Hubble flow. For relatively ‘close’ galaxies within 5-10 Gly the effect is somewhat marginal but close to the causal horizon this effect dominates. This means that if probes don’t have substantial onboard methods of acceleration, then they must be launched at substantially higher velocities such as 0.99c or 0.999c to arrive within a feasible timeframe. This obviously massively increases the energetic costs of launch and the amount of shielding required, as well as stretching out total flight times much longer. The shielding is probably the principal problem since at 0.99c the probe experiences the forward-facing universe essentially as a particle accelerator beam and any dust impacts, even micron-scale, have immense energy.

Reachability and Retrievability Horizons

Now that we have dispatched the probes, it is worth talking about two extremely important volumes. Firstly, there is the ultimate causal horizon — the volume of the universe that our probes can theoretically reach and start colonising. With probes at 0.9c this is approximately 15 billion light years in radius. Secondly, there is the useful retrieval radius, which is where we can not only reach in a single shot but where we can achieve sustained communication with and ultimately hope to potentially receive mass-energy sent back from those distantly colonised galaxies to the milky-way where we can use it.

Super naively, the outer limit of useful retrieval radius is half the maximum theoretical radius. This is the case where once a probe lands in the target galaxy it can immediately send a message back, but messages after that are lost. Half the radius or ~7.5 billion light years is the radius at which it is asymptotically possible for us to receive the probe’s return signal. For the entire rest of the causal horizon the probe’s mission is one way only. After a certain time in-flight we will never be able to communicate with the probe again.

Of course simply being able to exchange one incredibly delayed and red-shifted message from the probe is not enough. Ideally we want to receive a large amount of mass-energy back from the target galaxy. This narrows the constraints substantially. If we try to preserve a 1-2 billion year window in which we can meaningfully communicate or transmit energy then this gives us a proportionately lower radius.

This is a fuzzy boundary but I will assume that the rough radius that we can meaningfully retrieve a decent percentage of the mass energy from is approximately ~5 billion light years. Put together this means that the total mass energy in our causal-accessible universe we can theoretically accumulate back in our local galaxy comes to about 1-8% of the total energy causally accessible to us.

This means most of our ‘civilisation’ in terms of mass-energy will become causally cut off from us immediately upon or soon after ‘birth’, although if we do decide to send energy back to the Milky Way or our local group, that this will become by far the highest energy density region of the universe in our causal horizon.

So we would have to decide for the galaxies within the effectively retrievable horizon whether to let them branch off as ultimately causally separate civilisations using their own energy or whether we want them to become essentially energy transmitters to send back energy to the local group? This ultimately depends on your values — whether you care in an abstract, utilitarian sense about there being either copies off our civilisation or some kind of life/mind/flourishing occurring far away, even if they can never causally interact with you, or whether you prefer the energy to be sent right back to you so you can expand and increase the long-term energy consumption of your civilisation within your causally-connected local group.

For the galaxies beyond the retrievable horizon, this basically means whether you care about there being civilisation at all that is not causally connected to you. If yes, you should colonise, if no you should be indifferent whether to colonise or not (leaning no because of the small but extant cost of colonisation). My broad suspicion is that most ethics would say to colonise even if the resulting colony will not be able to communicate back to you or interact in any way.

The more arguable case is our indifference curve between sending back energy to our home systems vs letting independent civilisations flourish in these galaxies. If the galaxies are close to us, then we can beam back the energy very efficiently. However, as we approach the asymptotic horizon of retrievability, the efficiency of beaming the energy back decreases markedly. At the horizon the efficiency drops to zero, but even before that the beams we would be receiving become highly red-shifted, losing a good fraction of their energy, and arrive late in our time. This means that we face a continuous trade-off curve up until the horizon of retrievability. If we put no value on energy being used locally by causally disconnected systems, we should colonise right up until the horizon, even at extreme inefficiencies. If we care somewhat but not as much as having energy usable within our local group, then this places a point farther away from the horizon where we should stop instructing the probes to beam energy back to us and to just develop their own civilization.

However, it could be that if we are indifferent between having causally separated civilisations which originated from our local group, or our core local-group civilisation use energy, then we should never do any energy beaming back at all and solely use the resources of our local group. This is because the act of beaming and the level of energy extraction in the colonised galaxies necessary to support it is necessarily somewhat inefficient compared to just using the resources locally.

This means there are effectively two classes of probes, which may or may not be made differently. Firstly, there are ‘civilisation-building’ probes which are sent out with instructions to essentially develop independently on their own and use the local energy for their own purposes. Secondly there are ‘resource extraction’ probes which aim to rapidly build up infrastructure to optimise their galaxies for energy extraction and then construct the beaming infrastructure to beam the energy back home to a pre-specified location and time. Finally, in terms of speed, there will likely be different probe designs targeting different regions. For probes relatively close to us – i.e. within ~5-10 Gly — we can probably use more ‘relaxed’ probe designs which operate at 0.9c. If we want to get close to our theoretical cosmic horizon in the 15 Gly range we need probes at 0.99 or 0.999c. These require immensely more shielding, energy to accelerate, and actual difficulty of constructions. Most likely, our K2 civilization would choose to launch the easier colonization wave first and fill out the harder ones closest to the horizon later. Theoretically this is the wrong order, as you want to shoot as close as you can to the horizon first, but over timescales of a few thousand years the difference in ordering is small.

An interesting question is that some of these probes, especially to galaxies close to the horizon, will be arriving extremely late in the lifespan of the galaxy. For instance, if a probe is dispatched at 0.999c to a galaxy 15 Gly away it will arrive approximately 50 billion years later in the frame of the target galaxy. By this point, almost all star-formation and large stars of the target galaxy will be gone. There will only be a smaller remnant population of red-dwarfs. However, this is much less bad than it seems. Even though these later galaxies will have expended much of their potential output of starlight, it seems likely that our K3 probe will not primarily be extracting energy via Dyson spheres in the long-run, since this is inefficient and captures only a small percent of the mass-energy of the galaxy. Instead, energy extraction via spinning black holes is vastly more efficient at mass-energy conversion. By this metric, the entire stellar era of the galaxy only expends a few percent of the mass-energy that is extractable from the galaxy by artificial K3-style means. This means that even an extremely dim and old galaxy is very far from being ‘dead’ and has only lost a small fraction of its total percentage. So late colonization, even extremely late colonization is still highly worthwhile. One minor question is whether Dyson spheres and stars are necessary to bootstrap to K3 status in the first-place. This is possible, but even in this case, the smallest red-dwarfs are estimated to burn for potentially trillions of years, so it is extremely unlikely any probe would be arriving so late as to reach a galaxy with literally zero potential stars to dysonize.

In fact, surprisingly, the normal ‘life’ of galaxies only burns a relatively small fraction of their total mass-energy. Even many trillions of years later under normal galactic evolution the vast majority of their mass-energy will survive. It is only over the really long haul as the galaxy slowly collapses into the central black hole and large fractions of it are dynamically ejected does the real mass fraction begin to dwindle. This occurs long long after the rest of the universe becomes causally inaccessible, and hence long after the meaningful colonization window.

Flight and Deceleration

Anyhow, returning to the probes, which are now in-flight, let’s discuss two concrete problems. Firstly, the probes need to be able to steer. Obviously, our K2 civilization will launch them with the best information present at the time, however light-lag means this is woefully out of date especially for very distant galaxies. At the extreme case, imagine our launch a probe at a galaxy ten billion light years away. At launch time, our information is 10 billion years old. If the probe launches at 0.9c the arrival time to the galaxy is more than 11 billion years hence, so there is a total information delay of >21 billion years (in practice due to cosmological expansion the delay is much longer potentially up to 50 billion years!) between launch and arrival. There can clearly be no precise targeting at this scale. Obviously our K2 civilization will have much better cosmological models than today, but there is nevertheless irreducible uncertainty. When we launch, we will be launching at the vague region of predicted future mass-energy densities rather than specific stars or even specific galaxies.

What this means is that the probes need autonomous navigation and steering capabilities. I’m assuming the probes have advanced enough AI that the intelligence and autonomy aspect of this is not a problem, since later they need to autonomously regenerate an entire K3 civilization (!). The two main things you need for good navigation at a physical level are good sensors and good actuators. For sensors, building good enough telescopes is in theory fairly trivial for a K2. The issue is maintaining their operation over billions of years and hardening them against the onslaught of radiation that travelling at relativistic velocities entails. We assume, though, that this is a solvable engineering challenge given all the other engineering challenges needed to be overcome to build the probe.

A somewhat more challenging problem is the actuators. For the probe to be able to make large-scale course changes — e.g. to potentially be able to change its destination over potentially tens to hundreds of millions of light-years (which seems a sensible uncertainty given the tens of billions of years of light-lag), then this entails an extremely substantial delta-v. One advantage the probe has is that it does not need to make these changes all at the end. Rather, it can start adjusting course billions of years away, where over the course of a long flight even small initial velocity perturbations can substantially change the final destination. However, this does not entirely ameliorate the delta-v required for steering, which remains substantial. To be safe the delta-v is probably on the order of 10^4-10^5 km/s.

This is obviously large compared to conventional rockets (earth-escape delta-v is approximately 13 km/s) but not necessarily for a K2 probe when the delta-v doesn’t have to be exerted rapidly but can be slowly accumulated over millions-to-billions of years of flight time. There are two principal methods of doing this. The first is the most elegant, which is to use directed magnetic deflection to obtain lateral velocity changes. If we already have a magnetic shield we are deflecting charged particles around the probe. If we preferentially deflect particles opposite to our preferred direction, by conservation of momentum the probe gains lateral momentum in our preferred direction. At any given moment the force we get from this is miniscule, but compounded over millions of years this becomes substantial and is likely to be our primary steering option. Secondly, we can use extremely-high-Isp mass-driver or fission exhausts as effective rockets. We will likely want these as a fall-back in any case and to make more sudden manoeuvres if necessary (and they will certainly be necessary at the end of the later deceleration phase). If we can launch exhaust at 0.01-0.1c we can roughly budget about 3-10 tonnes of mass as propellant for these engines (or we can try to recycle some of the intergalactic mass as propellant). These requirements are challenging but are not showstoppers.

Once near the target galaxy, then comes the big challenge: deceleration. Our probe is travelling at 0.9-0.99c. If we want to do anything other than just whizz past the galaxy we need to decelerate, which means losing the immense energy that was spent back at home launching the probe in the first place. We cannot do this by standard rocket deceleration. The rocket equation would destroy us at these relativistic velocities. Instead, we need to use mass already at the destination to decelerate. The most obviously viable way to do this is via magnetic sail drag. Rather than having a relatively small magnetic shield built to deflect charged particles, as the probe ‘wakes up’ and begins deceleration at the destination, it can switch to a higher power source and deploy a large magnetosphere outside the probe itself. This can be created by powering on superconducting coils within the probe body itself. The goal of this field is to sweep up a much wider area of charged particles and deflect them, exerting braking force against the craft itself.

The good thing about this is that it does not require expelling propellant mass which must be carried, and that it can be precisely calibrated in extent to provide the desired braking g-forces. The probe is travelling through intergalactic space and can begin preparing to decelerate likely millions of years ahead of finally arriving at the destination. This means that it can make do with a relatively small bubble on the order of kilometres to tens-of-kilometers, and a relatively small g-force deceleration occurring over hundreds to thousands of light-years distance from the target galaxy. In intergalactic space the charged particle density is much lower so the field needs to be larger to achieve the desired deceleration. Once the probe is in the denser galactic halo, the density is much higher and the probe can shrink the size of the sail. Realistically the probe needs to shed most of its relativistic energy outside the galactic halo in order to not hit dense galactic regions at relativistic velocities which would likely be fatal.

The magnetic sail works extremely well for deceleration from relativistic speed. It works poorly for the final deceleration and approach since it works primarily by the relative velocity of the probe vs the charged particles, and once the probe is below relativistic velocities achieving the same deceleration requires an unrealistically large field. At this point, once the probe has reached ‘cruising speed’ within the galaxy, it has to switch to other means of propulsion. The final deceleration and target selection should likely evolve through multiple phases. Obviously magnetic sail into the galactic halo and to shed the core relativistic velocities. Then potentially an electric or photon sail during the target selection and stellar approach phase, and potentially to stellar capture. For in-system propulsion and ultimately landing, fusion or fission drives are a strong contender as well as potentially chemical rockets for final descent and/or tiny lateral corrections.

Building a galactic beamer civilization

Okay, so let’s assume the probes have arrived and decelerated and landed. Now they begin the fun work of trying to build up a K3 civilization from scratch. I won’t go into too many details here because I think this is, very obviously, far from settled science. Most likely this will begin by the initial probe landing on some kind of asteroid with good metallicity close-ish to a star to secure both power and easy access to a wide range of materials without having to deal with the atmosphere and gravity well of a planet. From there it will need to begin constructing other probes to slowly take over and convert its starting solar system into more manufacturing capacity. It will need to begin dysonizing its home star and then dispatch probes to all other regions of the galaxy and likely ultimately its local group, and thus begin the ascent to K3. With efficient technology this is likely possible within a few millions to tens of millions of years in the home galaxy and within a few hundred million years in the local group.

What then? Clearly this depends on the initial goal the probe was programmed with⁴. For now we will distinguish two broad types of goals. Firstly to build, using local resources, the kind of civilization that we would consider ‘good’, even if we do not have causal interactions with it. Hopefully this would involve uncounted trillions of beings experiencing some kind of joy and happiness, and all other good aspects we would endorse. All colonization efforts beyond the horizon of effective retrievability are necessarily of this type, and potentially some inside it. I am deliberately going to leave what goes on in these galaxies alone as they are beyond the scope of this post.

Secondly, we consider probes with the goal of simply disassembling their constituent galaxies and sending their mass-energy back ‘home’ somehow, so our originating civilization can use it. This is likely attractive to us if we care about causally connected regions to us having resources vs theoretical beings living far over the causal horizon having resources in the future. We will dive into this case in much more detail since we can directly engage with the physics required to send energy ‘home’⁵.

Let’s first dispense with a few bad methods. We cannot and should not attempt to directly send matter back home. We simply face the exact same situation we were in with launching the probe except dramatically worse, since we are arriving in the target galaxy billions of years later. Secondly, we cannot directly ‘move’ the galaxy itself back. Accelerating an entire galaxy up to the relativistic speeds required to actually return is a non-starter. Basically the only viable method is to send energy back as a beamed, highly focused laser, extremely similar to the intergalactic beam weapons we discussed previously, except that rather than designed as destructive weapons we design it as a medium for power transmission. This means that the beam area should be wider so the flux is in manageable ranges, and that the beam tracks a stationary receiver which is designed to handle its power and convert it into usable energy rather than just heat and destruction on the other end.

Now a key consideration is the timing. If the target galaxy is anywhere near the retrieval horizon, the probe is operating under a very strict clock. The longer it waits and the longer the beam operates, the lower its efficiency, which declines exponentially close to the horizon. This means that the colonising probe faces a very specific physics problem — how to convert as much of the mass-energy in the galaxy as possible into energy which can be beamed in a coherent collimated way to a distant target, and how to do this in the minimum possible time.

Firstly, the probe cannot simply convert all the stars to Dyson spheres and beam their energy back. This is bad for two reasons. Firstly, stars burn for too long — many billions of years. If the probe is operating near the horizon it has hundreds of millions to maybe a few billion years max. Secondly, the total starlight given off by a galaxy is only a small fraction of its total mass-energy. Starts do not perfectly fuse all their internal material. Even if they did, there is a huge amount of miscellaneous mass in planets, dust, and other objects throughout the galaxy. Finally, stellar fusion itself is inefficient even for the fused mass directly – converting only approximately 0.1% of the mass into capturable energy. We need a much faster and more efficient mass-energy converter.

Luckily we have one: black holes. In nature we see quasars which are incredibly bright objects, often emitting more energy than their entire host galaxy. Quasars are the accretion disk around large black holes. As matter falls into the black hole it forms an accretion disk similar to planetary rings. The insane conditions and immense friction caused by the infilling matter as it loses gravitational potential energy results in massive heating and ultimately enormously bright and energetic output. For non-spinning black holes the general estimate is a matter-energy of 5-10% of the infilling mass converted to radiation, which is surprisingly good already and vastly better than stellar fusion. For rapidly spinning black holes, such as Kerr holes, we can go higher to 27% in natural systems and 42% as the theoretical maximum and perhaps 30-40% as achievable. Outside of antimatter this is basically the best conversion ration we can get in known physics. Moreover, unlike stellar fusion in stars which is bottlenecked by their own size and density, black holes can eat through mass at a vastly higher rate. This makes controlled quasar construction and feeding with Kerr-style holes a highly attractive option for rapid strip-mining of galactic mass.

Effectively, what we would need to do is build a ‘farm’ of artificial quasars of roughly 10^8-10^10 solar masses. Build many of these spread out through the colonised galaxy so we don’t have to pool all of the energy in a single central location which means we can process the mass faster and expend less energy moving it. These quasars are fed with a constant stream of tens to hundreds of solar masses per year, ideally all entering the black hole with a high and carefully controlled angular velocity to spin up the hole. We then extract energy from the accretion disk and the spinning black hole system via the Li and Paczynski process. The theoretical maximum energy efficiency of this is 42% of the rest-mass can be converted into energy.

Naively, quasars emit extremely high energy and unpleasant radiation across a wide spectrum. Directly channeling this and sending it back is probably bad. Instead, we need to capture the resulting quasar radiation locally, convert it into electricity, then convert that electrical power into a beautifully collimated beam of uniform phase and frequency and beam that precisely controlled beam back to the home region. Ideally, the beam specifics would be agreed upon and programmed into the probe before launch. If all the light is of roughly uniform and known wavelength, then the receiver can design their optical receiving system to optimally process and extract energy from it.

Effectively, instead of building Dyson spheres round the stars, the plan is to star-lift the matter out of the stars, push it into artificial quasars, build Dyson spheres around these quasars, and then use the energy from these dysonized quasars to generate and collimate a ridiculously powerful beam aimed back at our home galaxy. Similar to the space warfare post, the aiming civilization has a problem that it is targeting using information billions of years out of date. Moreover, it is an open-loop problem. If the beaming galaxy beams to the wrong spot, there is no mechanism by which the receiving civilisation can let them know or give them feedback. This means that there must be relatively precise instructions given to the probe either at launch or shortly afterwards of the spacetime coordinates to which they must beam their energy billions of years in the future, and then this ‘contract’ must be upheld by both sides without causal contact. I.e. the beamer must actually beam at the correct time to the correct coordinates in the correct frequency etc without the original instructions becoming distorted or corrupted, and secondly, the receiving civilisation must actually build the receiving infrastructure in the correct place by the time the beam starts in a way optimized to actually receive the beam.

How much power is in this beam? With our aggressive Kerr-quasar setup we can probably reach 10^40-10^41 W of power? This is approximately the power of 26 trillion suns (this is much much greater than the normal brightness of a galaxy because firstly normal stars burn for many billions of years and we want to export the energy much faster in only 1-2 billion years and secondly because normal stellar fusion only converts a <1% of the mass energy of the galaxy while we are aiming to convert ~20-40%.

Naively this would just appear as a massive death ray, similar, if not more powerful than the intergalactic lasers we discussed in the space warfare post. However, the key is to spread the energy over a wider region at the receiver’s end. To make a death ray you tightly collimate the beam so it focuses on a tiny spot. To make a power transmission device you spread the power out over a wider area (this necessitates a much larger collector, but we are talking about a local-group-spanning civilisation as the receiver). To get a rough sense of scale, let’s assume we are transmitting at 10^40W. If we want the power density at the receiver to be of the same scale as sunlight on earth the beam (and our receiver) needs to be approximately 200 light years across. However, in practice clearly we would do higher power densities. If we can receive at 1MW/m^2 we only need a 6-light-year receiver radius and if we can receive at 1GW/m^2 then we need a receiver of only 0.2 light years radius. Clearly these are massive mega-structures, but easily within the capability of a K3 civilisation to construct.

From K3 to K4

Let’s think about things in a bit more detail on the receiving end. Firstly, they are not receiving just one beam. Even if their colonisation radius of which they expect to receive beamed energy is just 5 billion light-years in radius, this is still billions of galaxies, and the beams for many of them will overlap in time. This means that at the receiving end we don’t just have to build one receiving station we have to build billions of them. This is still possible for a K3 civilisation, but it imposes large logistical challenges. Not only must they tell one probe where to send their energy but they must handle the logistics of tens of billions of receivers and beaming zones, and they must know all of this ideally at the time they dispatch the probes in a way that is accurate 2-20 billions of years later. This is doable but is obviously somewhat challenging.

Building all of these advanced optical receiving systems, and heat sinks and coordinating them across probably millions of light-years, etc is a massive industrial challenge, but one which we assume that a K3 can handle. It does not require any obvious violations of known physics. There are a couple of other subtle problems. Firstly, momentum. These galactic lasers consists of a very large number of photons and these photons will transfer their momentum to the receiving apparatus. The amount of ‘push’ that a galactic power laser can exert is massive. Now theoretically, this should be fine since if we colonised as a uniform sphere, then we should be receiving beams roughly equally from all directions. Thus, the net momentum should be zero. In practice however the universe is lumpy, beams turn on and off at different times and have different powers, and all the beams likely won’t be send to the same place but different stations in the local civilisation cluster. This will introduce all kinds of fun shear forces and likely impart spin to the global gravitationally bound civilisation. Nevertheless, to a first approximation this looks like a manageable logistical/control problem of scheduling beans (billions of years in advance) so that they mostly cancel, and doing active stabilisation once the momentum starts getting out of hand.

Secondly, and more troubling is the simple gravitational problem. Mass and energy inevitably must gravitate, and we are talking about concentrating the mass/energy of billions of galaxies within a relatively small region. If we naively put all of this mass-energy inside the Milky Way we will effectively have increased the milky-way’s density billions of times. In a small enough density this will mean we turn ourselves into a black hole. Even if we do not do that, we still turn our civilization into an extremely heavily-gravitating relativistic region, which has its own issues.

This means that although we ideally want a highly dense core civilization for communication reasons, we do not want to put all of our stored energy into this region. Instead, we likely want a very dense civilisational core probably roughly the size of the milky-way or much smaller— i.e. hundreds-to-tens-of-thousands of light years across, which only uses the energy it needs to function, and then spread out our beam-receiving and energy processing and storage infrastructure into a large volume on the outside of this, probably expanding over hundreds of millions of light years. This will make the whole region dense, but not so dense that we start to see serious gravitational effects. Within this region, then there will have to be a network of energy transfer from the outer storage halo towards the inner dense core.

Finally, there is of course the question of what to do with all of that energy. It is unlikely we would want to use it immediately. It seems hard to figure out what even a local-group-spanning civilisation would want to do immediately with the mass-energy of billions of galaxies. Moreover, if we care about maximising total computational power across our civilisational lifespan it is a terrible idea to use it immediately. Instead the best thing to do is to store all but a tiny fraction of this energy, and then slowly spend down our stockpile when the universe is much colder and amenable to efficient computation due to the lower CMB temperature and the Landauer limit.

So then there is the question of how do we store the energy? Barring new crazy physics there are essentially three primary mechanisms:

1.) Spinning Kerr-type black holes

2.) Small black holes for hawking radiation

3.) Antimatter production and storage

For Kerr black holes, we essentially use the energy to impart spin to the hole. Once the black hole is spun up it will continue spinning essentially indefinitely (barring minor losses from gravitational wave emission). When we want to restore the energy we can spin it back down using variations on the Blandford-Znajek process.

On the positive side, this is relatively easy to construct, and is incredibly passive since the black hole just sits there spinning, and we have control of when we want to harvest the energy by spinning down the black holes. On the negative side, it is not that efficient since we can only retrieve a maximum of 29% of the energy we put in, and it requires starting with a bunch of black holes already or else creating them which decreases the efficiency further since their mass is entirely lost to us until, potentially, they start evaporating at extremely deep timescales.

For the hawking radiation black holes, the theoretical conversion efficiency is close to 100% since all the mass must eventually evaporate into photons which we can theoretically capture. However there are a couple of deep problems. Firstly, hawking black holes are not really batteries, instead they are continual emitters on a timer, more like active reactors than deep storage. Secondly, they decay on a fixed timescale. You cannot dynamically increase or decrease energy expenditure from a hawking black hole, instead you can only change the distribution of the sizes of the hawking black holes you make. This is bad for aestivation since you want long periods barely using any energy. Thirdly, the natural emissions from Hawking radiation, as far as we know, are primarily high energy gamma rays, neutrinos, gravitons, and antiparticles, none of which are particularly great for efficient collection. Finally, over time the power output of the hawking black hole will increase as it gets close to final evaporation and in its final moments the power output will be dramatically higher than the baseline. This potentially could cause issues with collection or damage the collection machinery if not properly handled. Theoretically you could avoid this explosion at the end by feeding the hawking black hole mass to keep its mass constant, however if you want fast decay lifetimes the event horizon radius is so small that this becomes practically difficult.

Another minor issue is that if you want hawking black holes that evaporate reasonably quickly on non-cosmic timescales, such as over a billion years, the hole itself can become incredibly small. This makes it hard to handle both because of its tiny size, incredibly power density, and strong gravitational and tidal effects. For instance, a standard schwarzchild black hole with a hawking evaporation timeframe of a billion years would require mass of order 10^10 but have an event horizon of 10^-16m — or smaller than the radius of a proton (!). Similarly, creating such a black hole would be incredibly difficult. The technology to actually concentrate the incredible energy to match the mass of 10^10kg within a femtometer diameter is nontrivial, to say the least. Certainly this cannot reasonably be done by collapsing or imploding ordinary matter, you would have to use some kind of ultra-precise intersection of ultra-high energy laser beams.

The final energy storage method is large-scale antimatter storage and annihilation. Anti-matter sounds somewhat crazy but has a lot of advantages although it also has large up-front costs (even compared to the other methods!). The general idea would be to use the incoming beam energy to power industrial scale particle accelerators to generate proton/anti-proton and electron/anti-electron pairs, then magnetically separate the matter and antimatter to produce large clouds of hydrogen and anti-hydrogen fuel.

Once separated, however, the antimatter and matter could be stored in an inert way, and could be reacted in controlled quantities whenever desired. You would not want to use active containment to store it, instead you would primarily store by distance. Keep your antimatter hydrogen clouds light year apart from your normal matter clouds, and use regular gravitational effects to keep the antimatter clumped together. When needed, you would need to build an antimatter infrastructure to take tiny parts of your antimatter cloud and move it into a controlled antimatter-matter reactor.

The primary issue here is that we need to build antimatter infrastructure to handle the antimatter clouds. Since we are storing it primarily as neutral anti-hydrogen, we cannot easily control it using electromagnetic means (nevertheless trying to control millions of solar masses worth of charged positrons or anti-electrons is completely doomed and would explode immediately from self-repulsive forces). Instead, we would need to create anti- versions of basically the same infrastructure we would use on the matter side to marshal and control these hydrogen clouds. This requires building a mini anti-industry. Anti-carbon, anti-iron, anti-silicon, anti-factories and anti-computers etc. This is theoretically fine. Anti-chemistry and anti-metallurgy should be the same as regular chemistry and regular metallurgy. However, to bootstrap this from pure anti hydrogen requires mastery of anti-fusion in industrial amounts. We assume this is possible for our advanced K3 civilization but we admit this is highly speculative physics.

In this scenario, there would basically be huge clouds of anti-matter with their own dedicated processing machinery and anti-computers to control it. This would be separated by light-years of hard vacuum from the actual matter. Due to gravitational clumping this setup would be stable without constant energy input, nor in particular danger of destabilization and massive annihilation. The regular computers and anti-computers can coordinate using standard photons since they behave normally with both matter and antimatter. When power is needed, the anti-industry can gather and dispatch pellets of anti hydrogen from the cloud and send it to the reactor location in exactly the same way that this can be done with regular matter. Then they can be reacted in a controlled way as desired.

Regular hydrogen-antihydrogen annihilation has an energy efficiency of just above 50% because the naive decay path produces pions that decay into neutrinos with a substantial fraction of their mass energy. However, an advanced K3 can likely figure out a much better annihilation pathway. For instance, even just arranging for the pions to be captured by regular matter and undergo strong nuclear reactions followed by regular radioactive decay rather than neutral decay converts most of their energy into gamma rays instead increasing theoretical efficiency to ~90%. Likely even better annihilation paths can be engineered by a serious K3. This means that the effective energetic efficiency of the matter-antimatter annihilation storage scheme is likely above 90% and thus effectively unmatched except by carefully fed hawking black holes.

So, to summarize, the upsides of antimatter are relatively stable storage, control over exactly how and when to extract the energy, and extreme efficiency of mass-energy conversion by annihilation. The downsides are needing to essentially bootstrap from scratch a pure anti-matter periodic table and industry from scratch by direct anti-fusion. Secondly, although gravitational clumping and construction of cold dense antimatter objects should enable passive storage, there is still always the risk of catastrophic failure if somehow a large amount of regular matter was to end up in the antimatter zones.

A real K3 would likely use some combination of these storage mechanisms. Probably initially the spinning Kerr black holes on the natural black holes while the antimatter storage system is being spun up and just for diversification. For known fixed energy usages across time, it is potentially sensible to construct a schedule of hawking-evaporating small black holes. Then for large scale passive energy storage, creating and managing huge clouds of anti hydrogen in completely isolated antimatter zones, with a mirrored full anti-matter industrial ecosystem would be the long term solution for the majority of the K3s longest-term energy storage needs.

The longest term

Okay, so let’s assume that our K3/K4 civilization has successfully intercepted all the beams and stored all the energy from tens of billions of galaxies. What does it do with it then? Obviously this is a highly speculative question and I can just gesture at some answers.

First, we assume that the broadest general goal of such a civilization is to maximize the total amount of coherent computation possible over time. This requires two things — firstly getting free energy to do computation with and secondly, having the computation be centralized within a small region for coherent communication. I think the second is under-appreciated. Light-lag imposes extremely strict limits on coherent K4-scale computation. This means that a K4 wants to centralise its computational core to be extremely dense to minimise communication time⁶.

Secondly, if we want to maximize compute, we want to wait for a very long time for the CMB to cool. This is because the Landauer limit sets the minimum energy for bit erasure as a direct function of the temperature. By waiting for temperature to decrease we gain massive order-of-magnitude gains that outweigh almost anything else. This means that a mature K3/K4 probably does the minimum amount of compute necessary and energy-expenditure to maintain itself after the energy collection/storage phase and just waits until everything is incredibly cold to begin computing.

An interesting point is that the centralisation issue becomes vastly less severe once we are computing close to the landauer limit. This is because heat dissipation becomes the primary obstacle. To keep temperature near the minimum, requires both massive passive radiative cooling apparatuses but also an extremely small amount of energy consumption, far less than would be required to power full Jupiter-brains. Computation thus becomes extremely serial and extremely slow. The ultimate landauer efficiency looks more like a laptop at the end of the universe than a galaxy-spanning computronium wave at the end of the universe. This idea of waiting for an extremely long time at the end of the universe is not new. It is known as the aestivation hypothesis. Here, Sandberg proposes that one of the reasons for the fermi paradox is that it is just much better for aliens to ‘aestivate’ — i.e. hibernate — waiting for the universe to cool down before starting computation because of the landauer limit. This is certainly true. By computing with CMB at 10^-20 rather than the approximately 3K it is today we gain a factor of 10^20 in computational efficiency. This is absurdly massive. At the same time, even with this humongous computational bounty, it is still better right now to spend energy accumulating the mass-energy of this universe into a storable state so we can use it much much later during these future epochs. As long as we are not spending more mass-energy than we gain right now it is valuable to do so. So mere aestivation does not resolve the Fermi paradox, it just means that once they have accumulated all the energy they can in the reachable universes, civilisations should begin serious aestivation at that point.

Once the CMB cools down sufficiently, they should then begin to compute extremely slowly and patiently, with massive radiators in a single incredibly dense computational region. What they do with that compute, who knows, but the scale of the compute is absolutely incomprehensible. Before the end they should be able to achieve at least 10^100 irreversible bit-erasures and potentially closer to 10^110 or slightly more. Presumably some of this compute is dedicated to figuring out if there are any loopholes in physics that allow them to go longer. Whether this is possible or not, who can say at this point?

So, to sum everything up, the broad strokes of the universal colonisation future looks as follows:

1.) Next few hundred to hundred thousand years: reach K2 civilisation status, construct our von-neumann probes and their launchers. Launch hundreds of billions of probes over a window of a few hundred thousand years while beginning colonisation of the rest of the milky-way and ultimately the local group.

2.) Over the next 1-2 billion years: Fully colonise the milky-way and local group. Become a K3 civilisation. Within a few billion years produce the necessary receiving infrastructure for the incoming beams you asked the probes to the closest galaxies to produce upon their dispatch.

3.) Next 2-20 billion years: receive all the energy beams from the retrievable causal horizon. Presumably store the vast majority of this energy into huge anti-matter clouds or into spinning black holes or small black holes for hawking radiation. Accumulate likely around ~10^62-10^66 J of stored mass-energy during this phase.

4.) 20 billion-1 trillion years: Mostly aestivate while waiting for the cosmic microwave background to cool down to below 10^-20-10^-30 kelvin.

5.) 10^12-10^32 years: slowly ‘spend down’ your accumulated mass-energy extremely slowly and coldly to maximise the amount of irreversible computation that you can do slowly as a civilisation⁷.

6.) Die either once you have spent your stored energy or else, if proton decay exists, at around 10^32-10^35 years when all of your baryonic matter infrastructure decays and you can no longer support any meaningful non-photon/black-hole-based infrastructure.

I think overall that is a pretty decent map of the far future. Fun times ahead!