Latest Discussions - COMSOL Forums

Rayleigh Benard Convergence in Square Cavity Enclosure

Fri, 27 Feb 2026 14:49:53 +0000

I am currently modelling 2D natural convection in a square cavity using the Boussinesq approximation and have encountered an issue when switching from a vertically differentially heated cavity (left wall hot, right wall cold) to a Rayleigh–Bénard configuration (bottom wall hot, top wall cold, vertical walls adiabatic). The fluid assigned is a nanofluid which the effective properties have been derived through the correct relationships and have been made in global parameters. Model setup: • Geometry: 2D square cavity • Physics: Laminar Flow (spf) + Heat Transfer in Fluids (ht) + Nonisothermal Flow • Coupling: Boussinesq approximation implemented via Volume force as no tick box for boussinesq approximation • Boundary conditions: – Bottom wall: isothermal hot – Top wall: isothermal cold – Vertical walls: adiabatic – All walls: no-slip • Density in Laminar Flow: constant (rho0) • Buoyancy implemented as a Volume Force: Fy = -rho0 * g_eff * beta * (ht.T - Tref) • Rayleigh number varied by defining: g_eff = Ra * nu * alpha / (beta * dT * L^3) and performing a parametric sweep over Ra. Issue: For all Rayleigh numbers tested (10^3 to 10^6): • Velocity magnitude remains the same across each Rayleigh number • Temperature field remains nearly linear (conduction profile) • Hot-wall heat flux does not change • Nusselt number remains exactly 1 across all Rayleigh values which is definitely incorrect All parameters (Ra, g_eff, buoyancy force magnitude) are confirmed to change between parametric steps via Global Evaluation. The buoyancy force expression is non-zero and applied to the fluid domain. Interestingly, when I revert to the vertically differentially heated configuration (left hot, right cold), the model behaves correctly: velocity develops and Nusselt number increases with Ra. This suggests the model is converging to the conduction branch in the Rayleigh–Bénard case for all Rayleigh numbers, even above the expected critical value.

I would appreciate guidance on best practice for obtaining the physically correct convective solution branch in this configuration. I believe all my physics are correct in my model and buoyancy is active and the above image shows the velocity plot for all Rayleigh values (it doesnt change). Below is my mph file of it and velocity plot picture which doenst change for any Rayleigh value - so definently incorrect.

Solid Mechanics: Rigid domain + applied moment rotating cylinder in agar block — final rotation vs torque ~10× off from expected

Thu, 26 Feb 2026 23:01:44 +0000

Hello,

I’m looking for advice on whether my modeling approach is correct and what I can do to improve it. To preface this, I have no FEM experience and I am trying to learn COMSOL on the fly to get this working.

The Goal: I’m trying to reproduce an experimental torque-rotation curve. The setup is a 3D agar gel block with a rigid body cylinder embedded inside it. I apply a moment to the cylinder so that it rotates within the gel, and I want to plot the final equilibrium rotation angle of the cylinder measured from its initial horizontal position versus the applied torque. The cylinder is rotating about an axis that is orthogonal to its main axis (so it rotates like a baton about its center, not spinning about its length).

Conceptually, a moment is applied, the cylinder rotates, the gel resists the motion, and the system reaches an equilibrium rotation angle.

The issue I’m having is that the final rotation of the cylinder is about an order of magnitude off (~10x) from the experimental results, and I’m not sure why. The current rotation is very small compared to what is expected.

Geometry: I created a 3D block with a cylinder centered inside it. The geometry is finalized using Form Union. I have tried using Form Assembly but am struggling to get the model to converge. I believe Form Union may be a reasonable approximation since the gel is relatively compliant, so I assumed continuity between the cylinder and surrounding gel.

Materials: The gel block is modeled as a custom material representing agar, with Young’s modulus, Poisson’s ratio, and density taken from experimental values. The cylinder is modeled as a rigid body so I just list its density in the rigid domain settings.

Physics: I am using Solid Mechanics. The gel is modeled as a linear elastic material (I also tried a hyperelastic model, but it did not significantly change the magnitude of the results). The cylinder is modeled as a rigid domain (I have tried using a rigid connector, but I get the same results). I apply a moment to the cylinder about an axis orthogonal to its main axis (again, it rotates like a baton about its center, not spinning about its length). The rigid domain constraints and settings are: * Zero translation in all directions * Rotation constrained about z and x axis * Free rotation only about the axis where the moment is applied (y axis) * Rotation defined about the center of gravity Boundary conditions on the gel block: * Bottom face fixed * Side faces are rollers * Top face is free

Mesh: Currently using a coarse mesh, but refining the mesh does not appear to significantly affect the results.

Study: I have tried both stationary and time-dependent studies, and both give roughly the same final rotation.

Results: I am computing the rigid body rotation using a global variable evaluation. I have also plotted rigid body rotation versus applied torque and the rotation is way smaller in the simulation for all values of applied torque.

Any help as to how I can improve my model would be appreciated. The main questions I have are: is modeling the cylinder as a rigid domain inside a linear elastic (or hyperelastic) gel appropriate for capturing this behavior? What could cause the rotation to be off by ~10x? What should I check first to improve accuracy (mesh, boundary conditions, material model, constraints, etc.)?

I have attached my model and I can provide any additional information if needed.

Thanks.

Charged Particle Tracing: plotting beam size vs position

Thu, 26 Feb 2026 18:52:44 +0000

Hello,

The tutorial on charged particle tracing and also the webinar on "Simulating Charged Particle Optics" at 22:20 shows how one can plot for example the average particle distance from the propagation axis, i.e. rms beam size, as a plot of time. However in ion beam optics i am more interested in this property as a function of the position along the beam propagation axis. This means i want to evaluate the rms beam size at several xy-planes with different z-coordinates (if the z axis is the beam propagation direction). How could i do this, especially when i am using a particle release in a volume as initial condition and not a particle beam, potentially releasing particles at different times, potentially having some particles that will freeze to a wall and so on? Intersection Point 3D looks like the dataset i want, but only for a single plane. And if i use the additional parallel planes feature, then how could i separate the data from each of these parallel planes and know from which of the z-coordinates this comes, as they are only saved as one big dataset then?

Thanks in advance for any help!

PS: I attached an example model in which i would like to have this plotted.

Unphysical mixing

Wed, 25 Feb 2026 00:46:25 +0000

Hello everyone, I am getting unphysical mixing in a steady -state analysis, Transport of Diluted Species (TDS) model coupled with Laminar Flow. This exact same setup ran perfectly fine in COMSOL v6.2, but the issue occurs in v6.4. How can I adjust the settings in v6.4 to eliminate this unphysical mixing and get back my v6.2 results?

Unexpectedly high thermal stress

Tue, 24 Feb 2026 13:35:51 +0000

When temperature is increased from 293.15 K to 294.15 K, thermal stress increases by more than an 10 times magnitude. I have taken materials whose CTE values are reasonably close, but the stress is still extremely sensitive to temperature. I have also defined the temperature dependent properties of materials. Side walls and bottom surface of the sensor is taken as fixed constraint while cavity & diaphragm surface is taken as free constraint.

Any suggestions on debugging this kind of thermal-stress explosion would be greatly appreciated.

Transformer Simulation Issues – Clarification Required

Tue, 24 Feb 2026 06:33:21 +0000

I am simulating a transformer with primary and secondary coils wound around the core limb. During the simulation, I encountered the following issues:

1. Issue with B-H Curve and Flux Density • I have assigned the B-H curve as the material property for the core. • After running a stationary study, the maximum flux density (Bmax) is showing approximately 400 T. • This value is physically unrealistic, since ferrite materials typically saturate around 0.3–0.5 T.

Question: Why is the simulation showing 400 T? Where could I be making a mistake in the setup?

2. Open Circuit and Short Circuit Giving Same Results

• When I excite one coil (either with voltage or current), • And apply open-circuit or short-circuit conditions to the other coil, • The simulation results remain the same in both cases. This behavior is unexpected because: • In open-circuit condition, only magnetizing current should flow. • In short-circuit condition, high current limited by leakage inductance should flow.

*Question: Why are both cases producing identical results? What mistake in the modeling or boundary conditions could be causing this issue? *

Transferring Stress Between Sequential Studies

Sun, 22 Feb 2026 15:54:35 +0000

I am performing thermal stress analysis of layered materials using Joule Heating coupled with Solid Mechanics. After completing the first time dependent study, I run a second study where I use the final time step of the first study as the initial condition for the new analysis. The temperature field is transferred correctly. However, the stress values at the initial time step of the second study do not match the final stress results from the first study.

I understand that the transferred variable is displacement field (comp.u), and I suspect this may be the reason for the discrepancy. Is there a proper way to transfer the stress field directly between studies, or is there a recommended approach to ensure consistent stress initialization, for example by transferring the strain instead?

Downgrading my COMSOL file version to accommodate it into the older version

Thu, 19 Feb 2026 16:48:06 +0000

Hello everyone,

Do the different module like convectivity enhanced conductivity not present in the COMSOL 6.3 latest version?

I am currently working in the COMSOL version 6.3, I have 6.1 version in my other PC, I want to open my that file in old version. I am unable to change my version. Please help me.

Meshing performance issue for perforated plate with ~21,600 holes

Thu, 19 Feb 2026 07:49:20 +0000

I am trying to mesh a 3D perforated plate with ~21,600 small holes (diameter = 1. mm , pitch = 1.1 mm) for an electrostatic study. The mesh either stalls near 99% or produces low-quality elements in narrow regions. I found discussions in COMSOL Forum about meshing small gaps and holes causing issues (e.g., "Meshing issues - small gaps" and "The meshing don't go further than 99%"), but none address this scale (tens of thousands of perforations).

My questions:

Is there any recommended mesh strategy for highly repetitive perforations?
Should I use symmetry/periodicity or a unit cell approach?
How can I reduce mesh complexity without losing field accuracy?

Any advice on handling large scale perforated mesh in 3D would be appreciated.

Request for Trial access code

Thu, 19 Feb 2026 04:05:25 +0000

Hello,

I would like to request a trial access code for COMSOL Multiphysics to evaluate its capabilities for my research on Battery thermal management system. Please let me know if any additional information is required.

Thank you for your assistance.

Best regards, Ritik Singh (MTech at NIT Jalandhar)

Floating electrode in nano scale

Wed, 18 Feb 2026 07:44:43 +0000

We now try to describe the floating electrode in nano scale. However, the calculation is difficult to converge and the floating electrode does not work well.

When i modify the scale of the system(nm→mm), the calculation easily converges and the electrode work well (The law of conservation of electric current holds true).

How should i configure it to calculate accurately on a small scale?

How to extract the Nusselt number of a flow pass a 3D flat plate

Tue, 17 Feb 2026 17:41:40 +0000

Could someone guide me through how to extract the Nusselt number from a flow past a flat heated surface. I am aware that we need to define the variables, but not sure the correct expressions to call. Any help is much appreciated.

Thanks

Photodiode simulation

Tue, 17 Feb 2026 07:11:51 +0000

Dear Sir/Madam

I am facing issues in the simulation of pn junction photodiode having Schottky contact. The expected output is the rectifying behaviour however in the IV sweep, there is no leakage current. Kindly suggest solutions. I am trying to simulate in 2D geometry but I have attached the file for reference due to file size constraints.

Thanks

Elastic Strain energy in prestressed eigenfrequency study-inclusion of static prestress?

Wed, 11 Feb 2026 18:03:16 +0000

Hello,

I am performing an eigenfrequency analysis of a suspended mechanical structure using the Structural Mechanics Module. The model includes gravity, leading to a significant static prestress in some components.

I extract elastic strain energy using the built-in volume integral

I observe the following behavior:

For gravity-loaded (prestressed) eigenfrequency studies, the elastic strain energy differs significantly from the corresponding eigenfrequency study with gravity removed.

The extracted elastic strain energy does not converge monotonically with mesh refinement, even though eigenfrequencies and mode shapes do.

Directional differences (e.g., longitudinal vs transverse motion) appear in the prestressed case but disappear when gravity is removed.

My questions are:

In a prestressed eigenfrequency analysis, does the reported elastic strain energy include contributions from the static prestress field, or only from the linearized modal deformation?
Is non-monotonic convergence of elastic strain energy under mesh refinement expected when static prestress is present?
When gravity is removed, should the elastic strain energy of low-frequency rigid-body-like modes be expected to become isotropic due to loss of a preferred direction?
Is elastic strain energy a reliable quantity for evaluating modal energy in strongly prestressed, gravity-dominated modes?

Thank you in advance for your help!

Regarding the assembly options in the solver

Mon, 09 Feb 2026 08:02:13 +0000

Hello everyone! I'd like to know about what is the difference between "eliminated output" and "non-eliminated output" in assembly, and does it have any impact on obtaining the stiffness matrix and load vector? Thanks.

The construction and materials of geometric layers

Thu, 05 Feb 2026 11:44:12 +0000

How can layers be generated in the geometric construction of irregular three-dimensional structures, and how can materials be added more conveniently to different layers

Why are there no sharp current spikes in COMSOL official 1D DBD plasma argon example? How to obtain them?

Wed, 04 Feb 2026 12:21:23 +0000

Hi everyone,

I am studying the official COMSOL example of a 1D Dielectric Barrier Discharge(DBD) plasma argon discharge, and I noticed that the plotted terminal current waveform is relatively smooth and periodic.

official COMSOL example

In many DBD studies (both experiments and published simulations), the discharge current typically shows sharp spike-like peaks (microdischarge pulses) superimposed on the displacement current background. However, in the COMSOL official 1D DBD argon example, the plotted terminal current appears to be a smooth periodic waveform without obvious sharp peaks.

I would like to ask two questions:

Why are sharp current spikes not visible in the COMSOL official example?
Which parameters/settings should be adjusted to observe sharp current spikes (if possible)?

Any explanation or suggestions would be greatly appreciated. Thank you!

Error

Mon, 02 Feb 2026 12:59:26 +0000

Error while computing in study what is this error about ' Feature: Stationary Solver 1 (sol1/s1) Undefined value found. - Detail: Undefined value found in the equation residual vector. There are 2 degrees of freedom giving NaN/Inf in the vector for the variable comp1.E10. at coordinates: (0.03179,-5.89722e-05), (0.03179,-0.000131553), ...' How to fix this ?

This is the link to the model. https://drive.google.com/file/d/1qNwZuvo4uazMab8wayqjtYDL1X7hpYcj/view?usp=drive_link

Improve the calculation speed

Sat, 31 Jan 2026 03:46:07 +0000

How to improve the calculation speed (convergence is very slow) for complex electromagnetic field models in COMSOL?

Unexpected Slowdown in COMSOL on Threadripper PRO 9985WX

Fri, 30 Jan 2026 04:40:28 +0000

We upgraded our COMSOL workstation CPU from an AMD Ryzen Threadripper PRO 7985WX to an AMD Ryzen Threadripper PRO 9985WX. Both systems have 256 GB of RAM.

On paper, the 9985WX is a newer and higher-tier CPU, but in COMSOL we’re seeing the opposite: performance feels clearly worse.

What we’re experiencing in COMSOL

• Geometry operations feel noticeably slower (changing geometry, rebuilding, etc. takes longer than before).

• During a solve, the UI and model interaction stutters / becomes laggy much more than it did on the 7985WX.

• Overall, the 9985WX system feels less responsive in COMSOL than the 7985WX system. CPU core utilization behavior looks different

We monitored CPU core activity:

• On the 7985WX, COMSOL tends to use the cores in a more uniform / evenly distributed way.

• On the 9985WX, COMSOL appears to leave several cores almost idle.

To rule out a hardware issue: • We applied external CPU stress/load tests and confirmed that all cores do run and boost properly under load.

• We also checked the rest of the hardware and did not find any obvious faults.

Thread/core settings tested (no improvement)

We tried changing COMSOL’s Number of cores setting:

• Using the default value.

• Manually setting values across a wide range: 48 up to 128.

None of these settings resolved the issue.

Memory behavior

Even though both machines have 256 GB RAM, we also observed that the 9985WX system uses noticeably more memory for the same work compared to the 7985WX system.

We request a solution or recommended configuration to restore expected COMSOL performance on Threadripper PRO 9985WX.