Blog

Technical articles on adjoint differentiation, quantitative finance, energy optimisation, insurance risk, and scientific computing.

How-To 5 min read

AAD Greeks don't match bump-and-revalue: the usual causes

When an adjoint sensitivity and a bumped one disagree, the cause is usually one of five: the bump size, a kink, a branch taken at the input point, an input that became a constant, or Monte Carlo noise. Each leaves a different signature, and each can be told apart in a few minutes.

Natalija Karpichina
How-To 4 min read

A derivative that jumps at a branch or a kink

At an if-statement or a max(), AAD returns the derivative of the side the code took. That is exact for the code that ran and silent about the jump next door. How to see which decisions a price depends on, how close each one was to flipping, and what to do about it.

Natalija Karpichina
How-To 4 min read

Adding AAD to an existing Python pricer without rewriting it

Record the pricer once with its market inputs marked, and one reverse pass gives every first-order Greek exactly; replay the same record at new inputs without running the Python again. What has to change in the code, and what does not.

Natalija Karpichina
A coding agent listing the decisions closest to flipping in a traced QuantLib pricing, each with its margin and source line
How-To 4 min read

Ask your coding agent where your Greeks come from

A local MCP server gives a coding agent the record of a QuantLib pricing instead of its source code. It answers which constants entered the price, which decisions nearly went the other way, which line computed the number, and what the price is at a new spot, exactly.

Natalija Karpichina
Benchmarks 3 min read

Same QuantLib source, seven compiler builds: which prices change, and why

We built QuantLib 1.41 seven ways and priced the same book through each. Compiler vendor and version change nothing. Fused multiply-add moves sixteen prices at the twelfth decimal. An ill-conditioned calibration moves in the fourth significant figure. Pick any two builds and see.

Natalija Karpichina
The Autocallable Risk Gap: Bumping vs Adjoint Greeks
Quantitative Finance 3 min read

The Greeks Your Risk System Isn't Showing You on Autocallables

Structured products with digital payoffs have a dirty secret: the Greeks your risk system reports are probably wrong. Bump-and-revalue on discontinuous payoffs produces numbers that depend on your bump size as much as on the market.

Natalija Karpichina
22x Faster Variable Annuity Risk: Breaking the Actuarial Bottleneck
Insurance & Pensions 5 min read

22× Faster Variable Annuity Risk

Every actuary says you can't apply adjoint differentiation to Variable Annuities: the ratchets are discontinuous, the lapse function is a step function, the withdrawal logic is a cascade of branches. We solved all of them.

Natalija Karpichina
Breaking the ALM Sensitivity Bottleneck
Insurance & Pensions 6 min read

Why Insurance ALM Still Uses a Method from the 1990s

Banks solved the sensitivity bottleneck over a decade ago with adjoint differentiation. Insurance hasn't. The barriers are engineering, not math, and none of them turned out to be showstoppers.

Natalija Karpichina
The Computation Bottleneck: Moving Beyond Data in Climate Risk
Quantitative Finance 6 min read

Climate Risk Is Not a Data Problem

The climate risk industry is built on a premise: banks get climate risk wrong because they lack data. The real bottleneck is computation: revaluing a portfolio under each scenario takes days, making sensitivity analysis infeasible.

Natalija Karpichina
The Differentiable Simulation Revolution: Beyond Reinforcement Learning
Machine Learning 5 min read

When Simulation Meets Neural Networks

Put a neural policy on the same differentiable tape as your simulation. Train with exact gradients. 50-350× faster than reinforcement learning, with better results and free Greeks.

Natalija Karpichina
The Anatomy of AAD-Native Architecture: Three Kernels and Four Abstractions
Architecture 5 min read

What AAD-Native Architecture Actually Looks Like

If you're designing a risk system from scratch, what should the architecture look like when you design around differentiation from day one? Four abstractions, three kernels, and one principle: separate the what from the how.

Natalija Karpichina
AAD Migration: From The 30-Year Rewrite to Incremental Upgrade
Quantitative Finance 4 min read

How to Actually Upgrade to AAD

Most AAD migration attempts stall because they try to refactor the entire library before turning AAD on. There's a faster path: change double to idouble, one pricer at a time, and see improvement at every step.

Natalija Karpichina
FRTB: Solving the Regulatory Compute Crisis
Quantitative Finance 1 min read

FRTB Isn't a Modelling Problem. It's a Compute Problem.

FRTB-IMA runs over weekends at many banks. The bottleneck is Greeks: bump-and-revalue makes sensitivity computation O(N) in risk factors. Adjoint differentiation collapses that to a single backward pass.

Natalija Karpichina
CPU vs GPU: Decoding the Quant Performance Gap
How-To 3 min read

Why You Might Not Need That GPU

The assumption that GPUs are 1000× faster than CPUs is wrong for most quant workloads. Modern CPUs with AVX-512 and adjoint differentiation close the gap, at 30% lower infrastructure cost.

Dmitri Goloubentsev
Python-Based Live Risk Platform
Case Study 1 min read

Python-Based Live Risk Platform

Sub-second portfolio updates in Python: AADC's JIT compilation eliminates the traditional choice between Python's development speed and C++ production performance.

Dmitri Goloubentsev
Why ML Frameworks Fail the Quant Finance Test
Machine Learning 2 min read

40-64× Faster Than JAX, PyTorch, and TensorFlow

ML frameworks are general-purpose differentiable programming tools. For quantitative finance workloads (Monte Carlo, path-dependent payoffs, iterative calibration), a domain-specific approach is 40-64× faster.

Dmitri Goloubentsev
From Batch to Real-Time Risk with Automated IFT
Architecture 3 min read

From Batch Risk to Real-Time Live Risk with Automated IFT

Overnight batch risk processing is the status quo. The Automatic Implicit Function Theorem turns any calibrated model into a live risk engine: 20ms pricing + Greeks for 1000 IR swaps via QuantLib.

Dmitri Goloubentsev
How-To 1 min read

How to Accelerate Existing Quant Libraries with AADC

A practical guide to integrating AADC into existing quantitative libraries, using ORE as the reference example, showing the black-box recording pattern that works on any C++ pricing library.

Dmitri Goloubentsev
Benchmarks 1 min read

Speeding Up QuantLib with AADC: ~300× Faster

Independent blog post by Serg Kozyarchuk demonstrating a 300× speedup on a 10,000 OIS swap portfolio using AADC with QuantLib, no QuantLib source code modifications required.

Dmitri Goloubentsev
How-To 1 min read

Accelerate QuantLib 6-100× Easily

Step-by-step guide to accelerating QuantLib with AADC: less than 1% code changes, 6-100× performance improvement, full adjoint Greeks included.

Dmitri Goloubentsev
Case Study 3 min read

ORE LiveRisk: 1 Million FX Trades Priced in 0.4 Seconds

A case study in accelerating Open Risk Engine without understanding its C++ codebase. The black-box approach: record QuantLib/ORE through operator overloading, compile, replay 245× faster.

Dmitri Goloubentsev
Python at C++ Speed: Real-Time Risk with AADC
How-To 1 min read

Accelerate Python Models with AADC

Prototype in Python, achieve C++ performance. AADC's JIT compiler extracts the computation graph from Python code and compiles it to native AVX-optimised machine code.

Dmitri Goloubentsev
Benchmarks 1 min read

Accelerating Simulations and Backpropagation with AADC

Joint Intel/MatLogica whitepaper documenting AADC performance on Intel Xeon CPUs: DAG extraction, JIT compilation, and benchmark results for financial simulation workloads.

Dmitri Goloubentsev
Quantitative Finance 1 min read

Accelerating ISDA SIMM Margin Optimisation with AADC

Application note comparing AADC, PyTorch, and NumPy finite differences for ISDA SIMM margin optimisation: AADC delivers exact gradients orders of magnitude faster.

Dmitri Goloubentsev

Interested in these opportunities?

Let's arrange a free demo for you and your team.

Book a Demo