The Mechanical APDL application no longer writes duplicated tabular load data across load steps to *.ldhi files when the data remains unchanged. This reduces file size and improves efficiency. During restarts, the application reads the necessary data from the *.rdb file, ensuring continuity without requiring repeated table entries. In previous releases, duplicated tables were written to *.ldhi files in every load step.

The automatic hybrid parallel feature, also known as auto-hybrid, now prevents hardware oversubscription on large clusters that previously degraded performance. The following behavior changes are included:

Single-node runs at low core counts

Auto-hybrid now applies threads only to large domains when reducing core counts (for example, -np 12 becomes -np 4 with 8 threads). This change may alter the solution printout compared to previous releases. Previously, threads were distributed evenly across all domains, regardless of size.

Multi-node runs

Auto-hybrid activates only when all compute nodes have the same number of CPU cores (homogeneous configuration). Therefore, the number of domains and threads determined by auto-hybrid parallel heuristics may differ from previous releases. This change helps prevent hardware oversubscription and associated performance decline.

Oversubscription protection

A final configuration check now runs before execution. If oversubscription is detected, the program issues a fatal error and stops the run. To proceed, rerun the solution with the -nt 1 command-line argument, which disables the auto-hybrid feature. Previously, no such check existed, and oversubscription could lead to poor performance or instability.

Restart and downstream analysis behavior

When restarting or continuing an analysis, auto-hybrid restores the previous configuration of processes and threads. If a downstream analysis launches with more processes than the upstream analysis, auto-hybrid reduces the number of active processes to match the upstream configuration.

For example:

In previous releases, the harmonic analysis would reduce to -np 2 -nt 2. In Release 2026 R1, the harmonic analysis reduces to -np 2 and does not activate additional threads (-nt 1).

Recommendation: Launch downstream analyses with the same or fewer processes than the upstream analysis to avoid unexpected reductions.

MPC (KEYOPT(2) = 2) bonded (KEYOPT(12) =5 or 6) contact excluding initial geometrical penetration or gap and offset (KEYOPT(9) = 1) no longer reports PENE, GAP, and SLIDE in the element nodal solution, *.ldhi and *.cnd files.

Moment convergence is now checked at the pilot node of TARGE170 when target elements have rotational degrees of freedom (DOFs), improving accuracy. Previously, this check at the pilot nodes was ignored if there were no structural elements with rotational DOFs in the model.

The sparse solver is now used by default in harmonic balance method (HBM) analyses. This change is expected to improve overall performance, particularly for systems with many degrees of freedom. While the solver may introduce slight numerical differences in results, these are generally negligible and should not affect the accuracy or reliability of the analysis.

Output messages now report the elapsed time (wall clock time) since the start of the program, instead of CPU time. This change provides a more accurate and relevant measure of total runtime. It accounts for all delays, including those outside CPU execution, and avoids confusion caused by thread-level CPU time aggregation.

Example output:

***** ROUTINE COMPLETED *****  ELAPSED TIME =        3.461 
*** NOTE ***                   ELAPSED TIME =       12.982   TIME= 08:00:00 
*** WARNING ***                ELAPSED TIME =       11.432   TIME= 09:00:00
*** WARNING ***                ELAPSED TIME =       65.347   TIME= 10:00:00

In previous releases, the output used the label CP to indicate CPU time. It measured only the time the processor spent executing instructions, excluding time spent on I/O operations, network delays, GPU acceleration, and other non-CPU activities. Additionally, in shared memory parallel (SMP) mode, CPU time was summed across all threads and cores. This could result in misleadingly high values, making it appear that certain simulation steps took longer than they did.

The nonlinear solution monitor file (*.mntr) now includes an additional field between variable fields 1 (VARIAB 1) and 2 (VARIAB 2). This new field reports current memory usage (in GB) at each converged substep, enabling better tracking of memory consumption during nonlinear analysis.

In previous releases, the *.mntr file contained variable fields 1 through 4. With this change, fields 2 through 4 have shifted to become variable fields 3 through 5.

For more information, see MONITOR.

This release introduces a restart mechanism for all CG drivers to improve solver reliability. The enhancement may improve solutions for models that previously converged slowly (below 1.0E-6) or failed to converge. Some test cases may show a slight increase in iteration counts due to the additional cycle.

When the restart mechanism is triggered, the solver performs an extra cycle of iterations. This second cycle starts from the most accurate solution found in the first cycle. Both cycles use the same maximum number of iterations.

Additional behavior changes includes:

In previous releases, when the solver reached the maximum number of iterations without meeting the tolerance, it would error out without completing the solution.

For more information, see Conjugate Gradient (CG) Solvers in the Basic Analysis Guide.

The PCG Lanczos eigensolver now uses a single stopping criterion for its internal conjugate gradient (CG) solvers. This criterion is based on the exact norm of the error and is expected to improve result accuracy. Some test cases might show a slight increase in the number of iterations.

In previous releases, internal CG solvers used multiple stopping criteria and stopped iterating when any one was met. Some of these criteria relied on approximate error norms.

for more information, see Conjugate Gradient (CG) Solvers in the Theory Reference.

Within a *PYTHON block, issuing mapdl.run(...), no longer displays output by default. Instead, it returns a string that can be printed by issuing print(mapdl.run(...)). To restore the previous behavior of automatically printing all output, set the environment variable MAPDL_PYTHON_WRITE_OUTPUT.

The updated impedance sheet formulation provides physically consistent behavior for low, matched, and high impedance values. When the sheet impedance is much lower than the characteristic impedance of the fluid (), the surface behaves as a soft or limp sheet, allowing pressure release. When the sheet impedance equals the characteristic impedance of the fluid (), the sheet behaves as an almost perfectly transparent interface, producing negligible reflection and allowing the acoustic field to propagate as if no internal surface exists. When the sheet impedance is much larger than the fluid characteristic impedance (), the sheet behaves as a rigid interface, strongly restricting normal velocity and introducing significant reflection. For details, see Impedance Sheet Approximation in the Theory Reference and Impedance Sheet in the Acoustic Analysis Guide.

The Ansys Product Improvement Program (APIP) now automatically enrolls participants. In previous releases, enrollment was off by default and required manual opt-in. You can opt out at any time by following the steps described here: The Ansys Product Improvement Program in the Operations Guide.

For non-isothermal loading, the backstress is now scaled differently. It uses only material parameter C in the nonlinear kinematic hardening model, resulting in changes to stress and strain behavior. In previous releases, non-physical results were observed, such as negative values for plastic work per volume.

For the contact elements, CONTA172, CONTA174, CONTA175, CONTA177, the following has been undocumented:

Instead, use the real constant CNOF and CNCHECK, ADJUST.

The following connection commands used to import early versions of CATIA (V4 and earlier) or ACIS files into Mechanical APDL have been archived and are considered obsolete. Instead, use one of the other products like the Mechanical application or Discovery or use one of the CAD/CAE packages supported in Mechanical APDL (see Introduction to the Connection Functionality in the Connection User's Guide).