HJARL AS · OSLO, NORWAY

Optimal control and coordination for physical systems

Differentiable simulators and learned policies for physical systems where linear proxies leave value on the table.

PILOTS & SHOWCASES:
1 • WHY - THE BOTTLENECK

Real physics isn't a smooth convex bowl.

THE CLASSICAL SOLUTION · MPC / SDDP

Finds the exact solution to an imperfect representation - linearizing physics and imposing strict rules on dynamics, distributions, and coupled systems.

THE HJARL APPROACH · NON-LINEAR PHYSICS

Finds an imperfect solution to a near-exact representation - complementary to classical solvers, improvable through iterative training, with a testable optimum.

TODAY · TRIANGULATION ON CONVEX PROXY
HJARL · EXPLORATION OF REAL SURFACE
Exact solution to an imperfect representationlocal minimumx* (π*, λ*)Imperfect solution to a near-exact representation
2 • WHAT - DISTRIBUTED OPTIMAL CONTROL

Local control over time, coordinated by signals in space.

OPTIMAL CONTROL · LOCAL DOMAINS OVER TIME

Each policy optimizes control in its local domain over time and within hard limits.

OPTIMAL SIGNALS · COORDINATION IN SPACE (IN R&D)

Shadow price signals pass up and down the hierarchy so local decisions align with the global optimum.

● agents optimize control locally
● coordination done over shadow prices
3 • HOW - THE HJARL TRINITY

Modular physics simulation with differentiability at its core.

∇f • Differentiable simulator

Modular physics components run forward while exact gradients pass backward through time, replacing trial-and-error sampling.

π* • Learned decision policy

Maps live system states and external drivers to control setpoints. Once trained, inference is immediate.

λ* • Equilibrium solver

Updates control policies and constraint shadow prices (λ*) together so the system converges within hard physical limits.

NOT A BLACK BOXEvery decision is traceable through exact gradients and shadow prices.
● causal sampling (π* → ∇f → λ*)● causal learning (∇f → π* → λ*)
∇fπ*λ*
4 • FOR WHOM - NOT JUST FOR OPERATORS

For live advice, performance benchmarks, and system design.

1 • Advise live operations

Read live state tags and return recommended setpoints and sensitivities as operating conditions shift.

2 • Benchmark operator and policy

Where the operator outperforms the policy, refine the simulator; where the policy outperforms, improve operating procedures.

3 • Co-optimize designs and policies

Treat physical design parameters as trainable variables alongside the control policy to size and configure systems before committing capital.

operator > agent → refine simulatoragent > operator → upgrade procedure
5 • WHEN - PATH TO RUNNING LIVE

From simulator to sandboxed advisor.

Example: Removr / Honeywell direct air capture pilot.

1 • Model the physics

Assembled a transient thermodynamic simulator with adsorption hysteresis and kinetics transients.

2 • Train in parallel

Trained the policy for 400M simulated hours (10,000 × 40,000h batches) across weather, power price, and CO₂ price regimes.

3 • Benchmark and run in a sandbox

Demonstrated meaningful economic uplift against the default operating policy, then deployed on a sandboxed server advising the Honeywell control system.

6 • WHERE - IN THE FIELD

Tested on physical systems that can be differentiated.

PILOTS & SHOWCASES
🌊 Hydropower1
Pilot • Skagerak Energi
💨 Direct air capture2
Pilot • Removr / SINTEF / Honeywell
⚡ AC power grids3
Live interactive showcase
🏭 District heating4
Proof of concept

🌊 Hydropower dispatch

Pilot • Skagerak EnergiOptimal control (>1y horizon)
The classical bottleneck

Convex proxy solvers (SDDP) must make simplifying assumptions on inflow distributions, price dynamics, and head dynamics.

Hjarl approach

Explores the non-convex surface across multiple end markets, arbitrary inflow volatility, and dynamic head. Extensible to load-based tariffs and coupled assets.

Key outputs

Discharge recommendations, sensitivities, and water values (∂V/∂W). Immediate inference.

∇f · simulatorModular waterway, generator, reservoir, and spillway simulator with arbitrary external driver distributions.
π* · policySets dispatch policy and ending state terminal value (whose gradients ∂V/∂W give reservoir water values).
λ* · equilibriumEnforces reservoir, flow, and power bounds.

💨 Direct air capture

Pilot • Removr / SINTEF / HoneywellOptimal control (thermodynamics)
The classical bottleneck

Default fan-speed rules optimize only for stability, ignoring how capture economics vary with temperature, humidity, and power prices.

Hjarl approach

Optimizes fan speed for profit across weather and market regimes while maintaining stability bounds on changeover frequency and heat exchanger icing.

Key outputs

0-100% fan speed from four live tags: intake temperature (°C), relative humidity (%), power price (NOK/kWh), and CO₂ price (NOK/kg).

∇f · simulatorTransient thermodynamic simulator with adsorption hysteresis and kinetics transients.
π* · policySets 0-100% fan speed, ending state value, and adaptive integration step.
λ* · equilibriumEnforces changeover frequency and heat exchanger icing bounds.

⚡ AC power grids

Live interactive showcaseOptimal signals (spatial coordination)
The classical bottleneck

Zonal markets and DC approximations fail to price the four AC bottlenecks: voltage, reactive power, active power, and line load.

Hjarl approach

AC power flow and equilibrium solver computing nodal shadow prices and sensitivities - built with general autograd rather than hardcoded derivatives, so it integrates directly with other optimizations.

Key outputs

Nodal shadow prices, congestion sensitivities (∂loss / ∂control), and AC-feasible dispatch.

∇f · simulatorAutograd-differentiable AC power flow (voltage, reactive/active power, line load) composable with other optimizations.
π* · policyDecentralized asset policies responding to nodal shadow-price signals (in R&D).
λ* · equilibriumDerives nodal shadow prices across voltage, reactive, and thermal limits.

🏭 District heating

Proof of conceptOptimal control (heat & power)
The classical bottleneck

Coordinating waste incinerators, oil/electric boilers, and storage buffers against volatile power prices and peak-capacity tariffs.

Hjarl approach

Minimizes fuel and peak-tariff costs while meeting heat demand, using storage buffers optimally across daily and seasonal cycles.

Key outputs

Boiler and storage dispatch strategies.

∇f · simulatorCoupled thermal grid, boilers, and storage buffers including peak-capacity tariff dynamics.
π* · policySets boiler and storage dispatch policy.
λ* · equilibriumEnforces heat transfer and supply bounds.
7 • IN BRIEF

The Hjarl thesis at a glance.

PIONEERING AI FOR SYSTEM CONTROL AND COORDINATION1 - 6
1 • WHY
Real physics isn't a smooth convex bowl.Classical solvers find the exact solution to an imperfect representation.Hjarl finds an imperfect, improvable solution to a near-exact representation.
2 • WHAT
Local control over time, coordinated by signals in space.Policies optimize each asset over time within hard operating limits,while shadow prices coordinate actions across bottlenecks.
3 • HOW
Modular physics simulation with differentiability at its core.The simulator runs physics forward and passes exact gradients backward,while an equilibrium solver updates policies and shadow prices together.
4 • FOR WHOM
For live advice, performance benchmarks, and system design.Advise or automate live operations and benchmark historical decisions.Train physical design parameters alongside control policies ahead of investments.
5 • WHEN
From simulator to sandboxed advisor.Assemble a system simulator, calibrate external drivers, and train a policy,then benchmark against historical operations before deploying in a sandbox.
6 • WHERE
Tested on physical systems that can be differentiated.Hydropower dispatch (Skagerak Energi) and direct air capture (Removr / Honeywell) pilots,plus AC power grid and district heating showcases.
WHAT PROBLEMS FIT

Systems that can be simulated, have dynamics, distributions, or couplings that break convex proxies, and use continuous inputs and outputs.

WHAT IS NEEDED TO START

A system walkthrough, historical data on external drivers, and a domain contact for Q&A.

SCOPE OF WORK

Simulator assembly, scenario calibration, policy training, and offline benchmarking.

FOUNDER · HENRIK BAKKEN — EXPERIENCE
2000-05
NTNU · M.Sc. Industrial Economics & Physics.Specialization in statistics and physics; master's thesis at the Norwegian Computing Center on copula theory (paper).
2005-08
The Boston Consulting Group.Consultant.
2008-14
SEB Enskilda Corporate Finance.Shipping & infra, director.
2014-19
Nordic Capital.Private equity, director.
THE PATH TO HJARL
1988
Gaming and tinkering.Got an Amiga in 1988.
2005
Thesis on dependency structures.Realized that causality was vital to learning.
2007
First differentiable simulator.Built to optimize gas flow in pipeline networks.
2015-19
Nighttime AI researcher.Built CNN in 2015 (ResNet), LM in 2018 (BERT), custom CUDA kernels in 2019.
2020
Founded Hjarl.A window to be part of solving learning.

* Optimal control is in pilot testing (hydropower, direct air capture); optimal signals is in R&D (AC grid showcase).