iiinova — where your digital self keeps evolving

Korbinian Kottmann @qottmann.bsky.social · 19/08/2026

While the lumper* in me doesnt mind a little lumping, it's great to know the actual difference between these techniques in a succinct fashion * en.wikipedia.org/wiki/Lumpers...

030

Korbinian Kottmann @qottmann.bsky.social · 21/03/2026

My impression is that hope is not very vague once you go beyond static things like ground states, at least I dont see how you'd do accurate dynamic simulations at scale Here is a sober perspective from arguably one of the top classical simulation groups quantumfrontiers.com/2026/03/12/t...

130

Korbinian Kottmann @qottmann.bsky.social · 02/12/2025

I will not tire of whining about overleaf and quantikz speeds

070

Korbinian Kottmann @qottmann.bsky.social · 13/11/2025

All I want for christmas is a (better) quantikz equivalent for @typst.app (and arxiv+journal support) Using overleaf.com after typst.app just feels like going around in a horse-drawn carriage in 2025 - gets you from A to B, and is kinda cool and vintage, but just extremely slow and cumbersome.

181

Korbinian Kottmann @qottmann.bsky.social · 04/09/2025

It is that time of year again! The @unitary.foundation qOSS survey is here! Please consider filling it out if you are using, developing or maintaining any quantum OSS like @pennylaneai.bsky.social or similar unitary.foundation/posts/2025_q...

073

Korbinian Kottmann @qottmann.bsky.social · 26/06/2025

I'm giving an (online) talk next Monday. Be there or be square 🟥 www.thewiser.org/agenda

162

Korbinian Kottmann @qottmann.bsky.social · 04/06/2025

Spotted: A photonic GKP state in the wild 🕵️‍♀️ so cool to see! www.nature.com/articles/s41...

051

Korbinian Kottmann @qottmann.bsky.social · 03/06/2025

Trying to understand fault tolerant architectures without getting into the nitty gritty details of surface code physics: The Game of Surface Codes 🐉 pennylane.ai/qml/demos/tu...

071

Korbinian Kottmann @qottmann.bsky.social · 28/03/2025

firefox has this, also possibility to select multiple tabs via shift or cmd and then close with cmd+W :)

020

Korbinian Kottmann @qottmann.bsky.social · 13/03/2025

In particular I find it pretty weird to contribute to such a news article underlining the story that dwaves simulations are spoofed (using laptops!) and then go "well we didnt actually simualte it in full, but we could, we just dont want to" 🤷🤷🤷

110

Korbinian Kottmann @qottmann.bsky.social · 13/03/2025

Yet, you get headlines like these, which I find weird at best www.newscientist.com/article/2471... In particular the part about a laptop beating dwaves experiments is pretty misleading as for TNs its just small scale experiments and in the other done using 4 GPUs (as written in the article itself)

310

Korbinian Kottmann @qottmann.bsky.social · 24/01/2025

when you accidentally drink non-decaf coffee and your brain is yelling every thought aaaaaaaaaa

040

Korbinian Kottmann @qottmann.bsky.social · 22/01/2025

Good day! You can seal the deal and delete the account altogether

110

Korbinian Kottmann @qottmann.bsky.social · 20/12/2024

Seeing our software and resources be used by other researchers and practitioners in the qc space gives me extreme "academic pride", for a lack of better words 🥲 Here's to more well-documented code in 2025

1121

Korbinian Kottmann @qottmann.bsky.social · 25/11/2024

Loved this! In particular this closing statement clicked for me :)

010

Korbinian Kottmann @qottmann.bsky.social · 19/10/2023

It's not perfect, but a great watch. I haven't thought that much about conservation of momentum since undergrad 🥲

320

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

We also run experiments on real quantum hardware, accessing OQC Lucy through Amazon Braket in PennyLane. 10/11

110

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

The advantage of ODEgen is that the resulting gradients are analytic. Comparing VQE runs with stochastic parameter shift (SPS) gradients, starting from the same initial values, we reach lower energies with ODEgen while using fewer quantum resources. 9/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

Finally, we decompose the effective generator into Pauli words and construct simple shift rules with two terms. 8/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

You may ask, what is the point to using a quantum computer in the first place here, then? As is common practice in quantum algorithms, we consider cases where a sufficiently expressive large ansatz circuit is constructed from multiple smaller but controllable constituents. 7/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

We use differentiable ordinary differential equation (ODE) solvers to compute both U and its gradient ∇U in a forward/backward pass. In this way, the complication due to the time dynamics is taken care of by classical simulation. 6/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

Instead, we look for so-called effective generators 𝓗 that encapsulate the derivative of the pulse gate U. But how do we obtain 𝓗? 5/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

One way to compute the gradients of such a pulse program is via the stochastic parameter shift rule. However, this involves a Monte Carlo Integration, which, on the other hand, necessitates executing quantum circuits for many intermediate times τ. 4/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

We are interested in pulse programs that compute the expectation value of some objective Hamiltonian (think, e.g., molecular Hamiltonian for VQE). For simplicity, let’s look at an example with a single pulse gate and drop the time dependence. 3/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

Many quantum computing systems can be controlled by pulses that steer the evolution via a time-dependent Hamiltonian. This pulse Hamiltonian can be parametrized by trainable parameters θ. Evolving for some start and end time results in a pulse gate U. 2/11

100

Korbinian Kottmann @qottmann.bsky.social · 02/10/2023

Are you tired of performing expensive Monte Carlo integration in order to optimize quantum pulse programs? 💸⚛️ Compute analytic gradients with ODEgen instead! scirate.com/arxiv/2309.1... A🧵1/11

140