Every claim our products make is backed by a paper. Methodology, convergence proofs and reproducible case studies — published openly. Comments and challenges to christopher@compoundingenergy.com.
Revenue forecasting for battery energy storage in liberalised electricity markets. The mathematical backbone of CE BESS Arbitrage and the foundation paper for the Compounding Energy stack.
BESS revenue forecasts that take wholesale prices as exogenous overstate revenue substantially in any market where the storage fleet is large enough to influence price formation. We frame the BESS revenue equilibrium as a fixed point of the composition of two operators — a price-formation operator P and a profit-maximising dispatch operator D — and prove existence under standard assumptions. The naive Picard iteration generically fails to converge in markets with steep scarcity tails; we show that Krasnoselski–Mann averaging restores convergence with explicit rate.
Variational-inequality formulation of the BESS cannibalisation equilibrium; Krasnoselski–Mann convergence theorem; reproducible stylised GB case study quantifying the price-taker bias at 100–300% (2-hour) and 200–800% (4-hour) at NESO 2030 fleet penetrations. ~22 pp.
Extends CE-WP-2026-01 from wholesale-only to the full five-product GB stack — energy arbitrage plus DC-L, DC-H, DR-L and DR-H. Per-product fleet commitment vectors, product-specific price-formation operators, and a single state-of-charge-coupled dispatch LP that allocates the battery across products simultaneously. Existence and Krasnoselski–Mann convergence under standard supply-curve and ancillary-clearing assumptions. Validated against published NGESO settlement data for the 2024–2025 GB BESS fleet. Headline finding: single-product naive sums overstate equilibrium fleet revenue by 35–90% at projected 2030 GB BESS penetration. ~10 pp.
Decomposition framework retaining nodal resolution, multi-stage stochastic structure and full N-1 security at GB scale. Combines Benders with Magnanti–Wong cut selection, level-bundle stabilisation, and an N-1 separation oracle producing contingency-aware optimality cuts on demand. Finite-convergence proofs for the integer master, geometric-rate bounds for the continuous master. Reproducible 30-zone GB case study. ~14 pp.
Unified framework for embedding non-linear physical models — aerodynamic wake interactions, distribution-feeder hosting capacity, conversion-efficiency surfaces — into MILPs via ReLU surrogates. Interval-bound propagation tightens per-neuron big-M values, often by orders of magnitude. Provably always-active and always-inactive neurons are pruned and replaced by their linear or constant equivalents. Worked 3×3 offshore wind farm case study with Jensen wake interaction.
Most operational energy-modelling stacks are trained against synthetic hindcasts — re-runs of the upstream weather, price or load model on historical inputs, with realised conditions hand-corrected back into training data. The result is calibration against a reconstruction, not against what production sees. Compounding Energy's position is the opposite: every shadow forecast customers see is itself the next training cycle's input. Training and production share an artefact, not a copy. This note explains why the discipline matters, what it buys in calibration drift, and what it costs to maintain. ~4 pp.
Each paper anchors a product in active build. CE-WP-2026-01 through 04 will be revised to v2 as journal-targeted versions (likely IEEE Transactions on Power Systems or Applied Energy) once external feedback is incorporated, expected late 2026 / early 2027.
The Compounding Energy paper library is shared by invitation only during private beta. If you have been issued an access passphrase, enter it below. Passphrases are rotated periodically; if yours stops working, please get back in touch.