Why we invested ReynKo
And how we think an investment in breakthrough theoretical framework might just be one of the most profoundly impactful climate investments.
“Turbulence is the most important unsolved problem of classical physics.”
– Richard Feynman (Nobel Prize, 1965)
This pre-amble is a bit unusual for our deal memos, but a breakthrough in turbulence is about the most complex physics and math most investors will ever encounter. In fact, its so unexpected that when thinking about climate pathways, turbulence and more generally fluids modeling doesn’t show up anywhere. We checked.
SOME CONTEXT
Lets start with current the state of the art in computational fluid dynamics. Scroll to the section on the computational cost of modeling turbulence based on current understanding of the governing physical equations. This computational challenge is a result of incomplete physical equations to describe fluid flow and ReynKo’s founder, Justin Beroz, has derived the missing math. His theoretical framework has been used to correctly predict data obtained from decades of experimental observations.
The immediate impact of this new theory is an improvement in modeling tools which in turn allows for a more rapid exploration (in software vs physical prototypes) of ideas to reduce turbulence and therefore increase efficiency, reduce energy usage, and improve resilience.
This opportunity is unique because we’re quite used to moving from promising prototypes, but we’ve never invested at a theoretic breakthrough stage. But in our view, we can move quickly from Justin’s theory to create classic disruption via faster/better/cheaper computer modeling. And historically better modeling environments quickly lead to to better physical implementations. In this case this could range from surface treatments of ship hulls or aircraft “skins” to active control systems for maintaining laminar fluid flows in pipelines or better weather models (yeah they’re simulating fluids in various states).
To go a step further, following are two great introductions to the state of the art in our fluid modeling capabilities.
Video explaining CFD strategies to reduce computational costs while mitigating inaccuracies resulting from missing math related to turbulence.
Video explaining the role of deep learning models to get some performance improvements in CFD simulations.
How do you diligence this?
It’s certainly helps to start with a person who has one MIT PhD and is about to receive a second. But that’s unfortunately not enough. We’re used to reaching out to prospective customers and so we did. The challenge is finding folks who are working on really demanding CFD problems. Fortunately we could get some baseline feedback confirming modeling/mitigating turbulence is a major pain point including founds working on problems ranging from aerospace to heat exchangers (yes, more fluids).
Then there is the math and the physics. The problem in this specific case is that the ideal reviewers will have deep knowledge of both fluids and math and so this left us a very, very small universe of folks to talk with. We’re early investors in Near Space Labs, which boasts rocket scientists who deal with turbulence as a part of their daily operational challenges, and could attest to the math around turbulence being hard to crack. Our community also includes black hole physicists who, while self-admittedly are not experts in turbulence, have written and read enough papers and understand the math enough to say “this doesn’t look like obvious gibberish”. Even after discussions, most still want to see the response from their peers (peer review process at work).
Ok, that’s the context. On the the analysis.
ANALYSIS
Potential Impact
Turbulent fluid flows may be responsible for about 30% of energy consumption in the US because it’s directly linked to drag in ships, planes and vehicles as well as fluid transport in pipes. More efficiency translates directly into increased range, which in turns means moving up the timeline for full electrification of long haul shipping and flight. Then there is also the potential to improve the accuracy of weather forecasts, which will yield better predictions for extreme natural hazard events like typhoons, hurricanes, cyclones, tornadoes, and floods.
In addition, an understanding of turbulent flow informs more resilient and sustainable static structures (buildings, infrastructure including wastewater treatment plans, climate-resilient structures, etc). Modeling wind loads more accurately can improve the material efficiency of designs, and increase a structure’s longevity. A building’s HVAC system can be designed to maximize passive flow and reduce energy required to heat/cool buildings.
Market Size
Market size is a function of how the company chooses to monetize. On one end are licensing business models which offer a chance to fastest revenue and wide impact, but captures the smallest slice of value. On the other end is the commercialization of computational models for specific problems like boundary layer solutions for aircraft that could result in new types of low drag materials and/or control systems.
Drag reduction translates into fuel and GHG savings today but also brings forward electrification because overall energy requirements are reduced, so electric drivetrains don’t need to reach parity with current dino juice options.
Example — one type of turbulent flow reduction are aircraft blended winglet efficiency gains. Alone they generate 4- and 6-percent fuel savings.
Example of design features, rerofits and other interventions to improve ship efficiency using current CFD tools.
Team
Justin Beroz CEO and founder. He developed previously unsolved mathematics that unlocks CFD advances. Christine Jacob (CJ) has deep industry and research experience in areas like materials and additive manufacturing. We can’t share much more about her work experience due to its secure nature . The two have been friends for 10 years and shared an office at MIT.
A lot of classical physics was defined decades ago. So for example, Reynolds number or mathematical approaches like Kolmogorov are well known ideas in fluid dynamics, but the people associated with these ideas didn’t build companies. However, they did build brands by virtue of having their names associated with core ideas in the field.
ReynKo has this possibility because Justin will be associated with a major development and this likely helps with early branding and co-branding for access to meetings and relationships that would otherwise be very hard for an early stage company. While CJs work is not public, there are benefits to having worked on hard problems in core areas related to national security, again purely from the perspective of attracting talent and/or getting access to prospective customers.
Approach
The team will develop a new generation of computation fluid dynamic software tools which can be licensed or directly owned and applied to build new opportunities. The new computation tools will take advantage of the mathematical insights that allow for vastly improved computation models and with that more rapid exploration of potential solutions to a wide range of applications that lead directly to better/faster/cheaper outcomes.
Basic vs full solver
Basic solver will focus on limited set of “simplest” applications that don’t include phase changes, heat transfer or can assume incompressible fluids. This excludes some applications like aerospace, heat exchangers, some aspects of climate models. But the team has a pathway to add these additional physics representations.
So one key question from customer discovery is whether the basic solver will find a good match with prospective customer applications or whether the solver will need to be built out further to enable the team to explore more applications.
New products will require investment beyond software such as building and testing physical prototypes and then setting up manufacturing, sales etc. In the short term, this is likely to mean joint development but as specific opportunities are better understood, ReynKo will develop and sell new hardware and control systems.
Applications
Beyond general solvers to implement the new math, there are likely to be specific models built to explore specific problem domains like increasing aircraft operating efficiency or improving performance of heat exchangers. In some cases, it may be desirable to create and model new designs that can be licensed or otherwise developed via joint ventures or entirely owned by ReynKo. This is a bit unusual for a traditional computer aided design software company, but given the potential scale and scope of the new modeling capabilities, it’s a real additional possibility.
In particular, it seems reasonable to expect that design software is one flavor, but control software might also be interesting to enable closed loop control in some applications. That would expand the TAM and likely speed up impact and adoption.
Distribution
Customer discovery work will determine go-to-market needs. For example, it’s likely possible to negotiate software licensing agreements with a few large software platforms like Solidworks, Autodesk, Ansys. However, in the case of specific applications such as designing new control systems for pipelines, it’s likely that a joint development agreement would be entered into.
Competition
Across the range of envisioned activities, ReynKo will encounter different types of competitors. Here are a few different lenses for competition:
- Do nothing — for some customers this might be the case because improvements are too marginal. This might be true for fluid modeling or specific application IP.
- CFD software developers — some will likely choose to build their own proprietary tools using what they find in peer-reviewed publications
- Internal teams at potential customers in industries like aerospace, HVAC, etc — might consider some CFD IP too important to outsource?
- Falling energy costs — might erode the value of efficiency improvements for say water treatment but have less impact on transport where energy storage costs and densities are unlikely to reach parity with fossil fuels for a while and even then, there are still incentives to drive down operating costs (see aircraft and ship examples)
- Other researchers — Who else is working on this, how does their framework compare, and are they planning on commercializing it? For example, some approaches might be less elegant mathematically but might still be good enough for some insights and applications. Georgia Institute of Technology developed a framework in 2022 (Source)
Other Risks
Most risks are related to how efficiently the team can move to paying customers. We have seen multiple similar approaches that allow teams to layer revenue such as fast licensing revenue for longer term, higher value capture, multi-year new product development. Additionally, affirming ownership and establishing protection strategy of IP while getting rigorous third-party scientific validation.
Does the math work? So far, yes. When it’s tested against a number of known experimental data sets, the agreement is excellent. The peer review process will ultimately provide more certainty as well ongoing evaluation against more test data.
Public good/knowledge vs proprietary information The mathematics will be published in peer reviewed journals like Science so it’s expected that the math will be understood and applied to existing CFD tools. But it’s likely that the new computational possibilities, a lot like improvements in LLMs, will lead to a large number of applications and testing of new ideas. Given the large universe of applications, we should expect that CFD will be a source of direct revenue but the largest opportunity for impact and revenue comes from solutions to specific problems.
For example, selling CFD to aircraft manufacturers is one likely path to market, but it’s likely that discovering new surface treatments will result in specific licensing opportunities as well as new physical embodiments and control systems.
There is a relatively small universe of people who will participate in the peer review process to verify claims that new math does in fact predict experimentally observed data.
Large search space means there are so many potential applications that the team will need to build a framework to prioritize, starting with more customer signal, which could take the form of LOIs to spell out assumptions and potential impact, economics and deal structures.
How long will it take to see widespread impact on new hardware? Ideally the value from new flow improvements can be retrofitted to speed adoption in existing vehicles or pipelines. One could imagine a vehicle wrap to change physical surface or the introduction of new firmware in pumping systems. Conversely larger changes might take a while to come to market because vehicles like ships and aircraft last decades before requiring replacement.
