NMR group · 2021
CPINN
Summer 2021 (May–July) project on CPINNs for NMR quantum pulse design, in Prof. Kavita Dorai's NMR Quantum Computing Group, IISER Mohali.
Research project (unpublished) · May – Jul 2021
Interactive abstract
Play with the idea
This interactive diagram needs JavaScript. The schematic and write-up below explain the same idea.
Layman view
In plain words
Draft — to be verified. Written in plain language from the paper and project notes on this site.
In nuclear magnetic resonance (NMR), the nuclei of atoms behave like tiny compass needles. Short bursts of radio waves, called pulses, turn those needles, and the exact shape of a pulse decides what operation it performs. Designing a pulse means choosing the radio signal that produces the wanted result.
A physics-informed neural network is a neural network that is trained to follow a known physical law as well as to fit examples. For pulse design, that law describes how the spins respond to the radio signal. In general, such a method can search for pulses without solving the equations again for every guess.
This page explains the idea in general terms. The project's own network, data and results are not described here until they are public.
Expert view
The work, step by step
-
01
Context
A summer 2021 (May–July) project in the NMR Quantum Computing Group at IISER Mohali, led by Prof. Kavita Dorai. The project studies CPINNs for NMR quantum pulse design.
-
02
Status
The code and report are still being prepared for publication. No results are shown here until they are public.
-
03
Method background (general)
In the rotating frame, a single spin driven by a radio-frequency pulse obeys dM/dt = ω × M with ω = (Ω_x, Ω_y, Δ). Ω_x and Ω_y come from the pulse amplitude and phase, and Δ is the offset from resonance. In quantum form, dU/dt = −iH(t)U with H = ½(Ω_x σx + Ω_y σy + Δ σz). Pulse design chooses the controls Ω(t) so that the final U(T), or the final state, matches a target.
A physics-informed neural network represents quantities such as the controls or the state as outputs of a network and adds the residual of the governing equation to its training loss, so that training pushes the outputs towards solutions of the physics. Which quantities this project's CPINN represents, and how, is not stated on this site yet.
Schematic
How it works
Contribution
My role
Publication