Three different bits of maths sit behind the plugins sold as physical modelling, and each one changes what comes out of your speakers.
Nobody prints this on the box. You get told about resonators, materials and exciters, while the underlying method stays buried in a manual footnote. It matters, because the method decides what the instrument can and cannot do
Modal synthesis, digital waveguides and finite difference time domain solvers all count as physical modelling. They differ in how much of the physics they genuinely compute. That gap turns up as CPU load, as tuning behaviour, and as how solid the low notes feel.
RipplerX gives you up to 64 partials per resonator, with two resonators you can couple in serial or parallel. That is a lot of tonal range from a structure that is, underneath, a stack of filters. The same architecture drives the dual-layer modal synth reFX released this summer, scaled up with more models and a proper effects chain.
Cost is the selling point. A modal engine runs comfortably on hardware that would choke on a full simulation, which is why almost every percussion-focused modelling plugin picks this route. The limit is that the model never computes what happens between one end of the object and the other.
The method models waves bouncing along a medium using delay lines, filters for energy loss, and nonlinear elements where the physics demands it. Efficiency comes from housekeeping. Losses spread through the medium get consolidated and calculated once at the end of a delay line instead of repeatedly along it.
The older Karplus-Strong algorithm turned out to be the same thing in miniature, a waveguide model of an ideal vibrating string. Those patents have now expired, which is a large part of why waveguide code keeps appearing in free plugins.
That buys detail you cannot get from a filter bank. You can excite the string at any point along its length, move the pickup to measure displacement somewhere else entirely, then sweep that pickup position with an LFO running up to audio rate. Drop the simulation accuracy and the tone goes softer and more inharmonic.
The bill arrives at the bottom of the keyboard. Lower notes mean longer strings, longer strings mean more points to compute, and the plugin caps out at ten voices, which it reduces automatically when your machine struggles. On weaker hardware, a low note can force it into monophonic behaviour.
It is a free string synth that will punish an older laptop, available as pay-what-you-want in VST3 and AU for macOS 10.13 or newer, Windows 10 and up, and 64-bit Linux. Version 1.1.1 arrived in July 2026. Run it alongside a modal plugin on the same part, and the difference in weight is obvious long before you look at the CPU meter.
Nobody prints this on the box. You get told about resonators, materials and exciters, while the underlying method stays buried in a manual footnote. It matters, because the method decides what the instrument can and cannot do
Modal synthesis, digital waveguides and finite difference time domain solvers all count as physical modelling. They differ in how much of the physics they genuinely compute. That gap turns up as CPU load, as tuning behaviour, and as how solid the low notes feel.
Modal synthesis stacks partials and stays cheap
The modal approach treats a vibrating object as a bank of damped harmonic oscillators, each sitting at its own frequency and dying away at its own rate. Change the damping and wood becomes metal. Nothing travels anywhere; it just rings and decays.RipplerX gives you up to 64 partials per resonator, with two resonators you can couple in serial or parallel. That is a lot of tonal range from a structure that is, underneath, a stack of filters. The same architecture drives the dual-layer modal synth reFX released this summer, scaled up with more models and a proper effects chain.
Cost is the selling point. A modal engine runs comfortably on hardware that would choke on a full simulation, which is why almost every percussion-focused modelling plugin picks this route. The limit is that the model never computes what happens between one end of the object and the other.
Waveguides came from Stanford and shipped in 1994
Julius O. Smith III coined the term digital waveguide synthesis and developed it at Stanford, and the university signed an agreement with Yamaha in 1989 to turn it into products. The VL1, VL1m, VL7 and the VP1 prototype all landed in 1994. The VL70m followed in 1996, and an EX5 workstation module in 1999.The method models waves bouncing along a medium using delay lines, filters for energy loss, and nonlinear elements where the physics demands it. Efficiency comes from housekeeping. Losses spread through the medium get consolidated and calculated once at the end of a delay line instead of repeatedly along it.
The older Karplus-Strong algorithm turned out to be the same thing in miniature, a waveguide model of an ideal vibrating string. Those patents have now expired, which is a large part of why waveguide code keeps appearing in free plugins.
FDTD solves the wave equation and eats your CPU
PartialString, built by Christian Baker at Different Instruments, takes the expensive road. It solves the one-dimensional wave equation in real time, holding the string in memory as a list of displacements and working out how each disturbance propagates along it.That buys detail you cannot get from a filter bank. You can excite the string at any point along its length, move the pickup to measure displacement somewhere else entirely, then sweep that pickup position with an LFO running up to audio rate. Drop the simulation accuracy and the tone goes softer and more inharmonic.
The bill arrives at the bottom of the keyboard. Lower notes mean longer strings, longer strings mean more points to compute, and the plugin caps out at ten voices, which it reduces automatically when your machine struggles. On weaker hardware, a low note can force it into monophonic behaviour.
It is a free string synth that will punish an older laptop, available as pay-what-you-want in VST3 and AU for macOS 10.13 or newer, Windows 10 and up, and 64-bit Linux. Version 1.1.1 arrived in July 2026. Run it alongside a modal plugin on the same part, and the difference in weight is obvious long before you look at the CPU meter.
