Visualization: A cognition amplifier



Computational codes produce huge quantities of numbers, but what scientists seek is insight. That is, they want to know what these numbers reveal about the phenomena under study. Visualization, transforming data into graphical structures, can enhance research effectiveness by leveraging human visual perceptual capabilities. Evolved over millions of years for survival, these capabilities enable us to see structures and perceive correlations and regularities around us. When properly visualized, we can apply these same perception capabilities, with little rational effort, to our data. In this tutorial, after analyzing a short definition of visualization, its major concepts will be seen in action using a popular chemistry visualization tool. The tutorial closes discussing paths for continued exploration of the future potential of these visualization techniques and tools. © 2013 Wiley Periodicals, Inc.

Seeing and Understanding

Eugene Wigner, looking at the results of a large quantum mechanics calculation, pleaded: “It is nice to know that the computer understands the problem. But I would like to understand it too.”[1] He did not want numbers or results. He wanted to know.

Computational codes produce plenty of numbers, but what scientists need is to understand them, what they mean, what they reveal about the simulated phenomena under study. To undertake this endeavor, scientists often rely only on their rational mind. However, they forget that an estimated one-third of the brain cortex neurons is devoted to visual perception and pattern recognition. Why not use that neurological machinery to comprehend numerical data? To make this possible, the numeric should be transformed into shapes and colors, the symbolic into the geometric. Essentially this is what visualization does.

This tutorial will define visualization in the context of quantum mechanics (QM) research. Hands-on exploration of a popular chemistry visualization software will overview tools currently available to researchers. A follow-on discussion will highlight paths for continued exploration, potential pitfalls, and the future potential of visualization tools that continue to improve at pace with accessible computing power.

Visualization to Amplify Cognition

“Visualization is the use of computer-supported, interactive, visual representations of data to amplify cognition.”[2] Hereafter this is the definition of visualization we use, but to understand its meaning, we start analyzing it here piece by piece.


Human beings have always created extensions for their bodies to overcome their limitations. Extensions are needed by their minds too because: “The power of the unaided individual mind is highly overrated. Without external aids, memory, thought, and reasoning are all constrained.”[3] This does not refer to raw computing power, but to the high-level processes, collectively called cognition, by which information is acquired, digested, and archived. Cognition that almost always is based on a mental model of what we are trying to comprehend.[4, 5] As these models are often visual,[6, 7] visualization facilitates building them for computational results by transforming numbers into images and graphical sketches of what we are studying.

Unfortunately, mental images are limited and ephemeral, so we need tools that offload those cognitive activities and anchor models to concrete artifacts, like pictures or physical mock-ups, to stabilize them. Not only that but also these cognition-supporting external representations contribute to mental models building and accordingly those models influence understanding.

Note that the definition says “computer-supported” and not “computer-based.” Visualization does not speak to machines, it speaks to humans. Human perception is the foundation on which visualization builds its effectiveness. Visualization is useless without human pattern recognition abilities and without our openness to creative and serendipitous discoveries.[8]

As a consequence, no automatic visualization builder exists. Visualization tools provide a toolbox of techniques and a palette of visual representations, but are the researchers, with their scientific questions and their knowledge of data, the ones who must use them to build effective visualizations and exploit these to foster discoveries.


Visual perception is never passive[9, 10]—our eyes are not TV cameras—so it requires vigilant attention to navigate, to move around the visualized graphical scene to build the correct mental model of what it represents. Think of a museum where you can look at a Greek statue only from its front side compared to another one where you can walk around it. Here you acquire a much better and complete understanding of the three-dimensional (3D) structure of this piece of art.

The same happens, for example, with the electron density values resulting from some QM computation. The researcher, through the visualization tool, accesses a set of two-dimensional screen images showing partial view of this volume, like a set of plane sections or volume-enclosing surfaces. Then, interacting with this visualization by changing plane, surface level, or viewpoint he mentally reassembles them into the essential parts of the volumetric structure.

Visual representation of data

Visualization taps into the capabilities of our visual system. A system evolved through millions of years in which survival relied on the ability to spot predators hidden in the shadows. Today we do not need to avoid wild animals, but we still immediately, without rational effort, see structures and perceive correlations and regularities around us. Visualization techniques work leveraging these capabilities, converting numerical values into graphical objects so features and patterns “pop-up” to our attention.[2, 8]

Do not assume that only 3D complex graphical representations are visualization; even a simple line chart is an effective visualization if it makes one think about something new and unexpected regarding data, as John Tukey observed: “The greatest value of a picture is when it forces us to notice what we never expected to see.”[11]

And what about scientific illustration? Seems it pursues the same goals as visualization. It provides understanding through sketches of reaction paths, simplified view of DNA helix or vivid representations of crystal symmetries. So what is the difference with visualization? One word only: data. Visualization provides a more direct representation of data compared to the averaged, idealized, or canonical representations that scientific illustrators most often produce for peer communication or education.

To amplify cognition

Purpose of visualization is enhancing cognition, igniting imagination, making visible the unseen behind the numbers, providing a holistic view of the data that the precision of the computational experiments often lead us to lose. We can affirm that purpose of visualization is to gain “insight,” not to make up pretty pictures, echoing the omnipresent Richard Hamming motto: “Purpose of computing is insight, not numbers.”[12] But what is insight? As Encyclopedia Britannica states: “It is the immediate and clear learning or understanding that takes place without overt trial-and-error testing. Insight occurs when people recognize relationships, or make novel associations between objects or actions, that can help them solve new problems.”

Unfortunately, the perceived role of visualization has gradually shifted from being a cognition amplifier to the subordinate role of a bag of techniques for creating nice pictures. In turn this change induced people to wrongly consider visualization as a luxury they can do without.

Visualization Hands-on Tutorial

To introduce the various capabilities common to almost all visualization tools, we take now a quick tour through the popular visual molecular dynamics (VMD)[13, 14] visualization tool and its functionalities. To make the example concrete we read and visualize a simple GAMESS[15] log file on a Windows laptop using the latest version of VMD (1.9.1), but everything could have been done on Mac or Linux with similar results.

The hands-on steps you can try are given below as numbered lists, while the rest of each section contains context information and supplementary material you can skip at a first reading.

Prepare the tool

  1. Download VMD from the VMD home page[14] and install it.
  2. Locate the EXAM01.LOG file. It is available as “Supplementary Material” to the article.
  3. Launch VMD in the machine specific way (for Windows this means clicking “Start > University of Illinois > VMD > VMD 1.9.1”).
  4. The VMD Console that announces the machine configuration (Fig. 1), the VMD Display and the VMD Main windows open (Fig. 2).
Figure 1.

During startup the VMD Console on a Windows laptop announces the hardware capabilities of the machine. Note the size of available memory and the presence of an accelerated OpenGL capable graphic card that provides also CUDA and GLSL rendering.

Figure 2.

Loading data in VMD. On the background is visible the VMD console with messages from the file reader. [Color figure can be viewed in the online issue, which is available at]

Read the data

  1. Go to the VMD Main window's menu “File > New Molecule…” to open the Molecule File Browser (Fig. 2).
  2. Push “Browse…” and select EXAM01.LOG as the file to open. Note the file format has been automatically detected.
  3. Open the “Determine file type” drop down menu to browse the list of file formats recognized by VMD. At the end select again GAMESS from the list.
  4. Push “Load” to read the file. During EXAM01.LOG loading the VMD Console lists messages from the reader that announces, among other things, that it added wavefunctions to timesteps 0 and 6.
  5. After loading the file, the VMD Main window appears as in Figure 2 where you see that the EXAM01.LOG file contains three atoms for a trajectory composed by seven steps (frames).
  6. Close the Molecule File Browser window.

VMD can load a long list of file formats, not all of them strictly structural formats. In fact VMD can load volumetric data (e.g., VASP_CHGCAR), geometries (like STL Stereolithography triangle mesh) or mixed volumetric and structural data (Gaussian Cube). QM results are saved mostly as GAMESS or GAUSSIAN[16] logs because these formats store specific information, like atomic basis set, normally not present in other ones. Also multiple files can be loaded together. In this case, all structures conserve their reciprocal positions.

In the cases above, the goal is to compare and relate different data to understand better their mutual relationship or to explain one data in the context of the other ones.

Navigate the graphical scene

  1. After loading the file, VMD displays the last timestep's structure as line structural representation to give an idea of what has been loaded without impacting performance (see Fig. 2).
  2. Click and drag the mouse on the VMD Display to rotate the molecule's representation.
  3. Use the mouse wheel to zoom the display.
  4. From the VMD Main window's menu “Mouse” change the mouse function to “Translate Mode.” Click and drag the mouse to pan the molecule.
  5. Change the mouse function by keying the one-letter shortcuts listed in the menu. Push 0 to select the “Query” mouse function. Click on one line of the molecule representation in the VMD Display window and read on the VMD Console (Fig. 1) the information related to the nearest picked atom.
  6. Return the mouse function to “Rotate Mode” pushing r.
  7. Using the buttons or the slider at the bottom of the VMD Main window move through the seven loaded timesteps.
  8. Return to timestep six.

You change the mouse function from rotate to translate mainly to explore a structure larger than the one used in this example.

Change structural representation

  1. Look how the Line structural representation fails to convey effectively the 3D molecule shape.
  2. On the VMD Main window go to the “Graphics > Representations…” menu to open the Graphical Representations window.
  3. Select the CPK representation, also known as ball & stick (see Fig. 3 top-left), from the “Drawing Method” menu. This is the most common way to show a chemical structure.
  4. Change to Licorice representation (see Fig. 3 top-right). If the visualized structure is big, then CPK shows too many details that obscure comprehension. In this case the Licorice method draws a more schematic structure but gives a perception of depth that is missing with the Line method.
  5. Change to VDW representation (see Fig. 3 bottom). This representation approximates the external shape of the chemical structure by drawing atoms as spheres of the corresponding Van der Waals radius.
  6. Visualize a more realistic molecule global shape changing to QuickSurf representation and setting “Grid Spacing” to 0.5.
Figure 3.

Selection of a representation for a molecule in VMD. The same molecule is rendered as CPK (top left), Licorice (top right) and VDW (bottom). [Color figure can be viewed in the online issue, which is available at]

Note that the bonds, if not present in the input file, are computed by VMD from interatomic distances.

Sometimes adding shapes or simplifying shapes helps understanding. This is not needed here because the loaded structure is so simple, but if you are visualizing a big biomolecule, rendering methods like Cartoon or Ribbons help. These representations simplify the molecular structure reducing it to the essential parts needed to understand a protein's structure.

Change colors and materials

  1. Set representation back to CPK. Colors in this representation have conventional meaning to distinguish chemical species.
  2. Change this color schema under VMD Main window's menu “Graphics > Colors…” selecting Type from “Categories” and H from “Names.”
  3. Change the color to 9 pink in the third column and see the hydrogen atoms changing color.
  4. Return to the Graphical Representations window and set representation to MSMS. Change “Sample Density” to 6.5. This representation shows the Solvent Excluded Surface for the molecule.
  5. Set “Coloring Method” to Mass.
  6. Rotate the surface to see the changing color on the surface of the molecule.
  7. On the Color Controls window select “Color Scale” tab and change the color scale to rainbow by selecting method RGB. The example has only two values to show; therefore, no difference appears in the visualized surface.
  8. Move the axis glyph to the upper-left corner of the VMD Display with the “Display > Axes” menu.
  9. Now add a color scale to show the mapping between mass and color (see Fig. 4). Push ''Extensions > Visualization > Color Scale Bar on the VMD Main window. On the Color Scale Bar window set “Autoscale” to On then push “Draw Color Scale Bar.”
  10. After looking at the colormap push “Delete Color Scale Bar.” Close the Color Scale Bar and the Color Controls windows.
  11. On the Graphical Representations window push “Create Rep.” It creates another representation for the same molecule. Change its type to CPK.
  12. The list below the button collects all the visualized representations. Select MSMS in the representations list.
  13. Change its “Material” to Transparent. The molecule becomes visible inside the surface (see Fig. 5).
  14. Push “Delete Rep” to remove the surface.
Figure 4.

The linear color map from Red to White on the right provides the viewer with a more direct and more correctly interpreted value for electron density compared to the traditional “rainbow” color scheme on the left that most molecular graphics packages use by default. [Color figure can be viewed in the online issue, which is available at]

Figure 5.

Orbital rendering using transparent surfaces. The visualized data is GAMESS EXAM01.LOG canonical wavefunction and orbital 6. [Color figure can be viewed in the online issue, which is available at]

Select atoms

  1. On the Graphical Representations window go to the “Selections” tab.
  2. Push “Reset” to clean the “Selected Atoms” field.
  3. Input name C in the “Selected Atoms” field then “Apply” and see the hydrogen atoms disappear.
  4. Push “Reset” to clean the “Selected Atoms” field.
  5. Double-click on all in the “Singlewords” list. Push “Apply” to show again the full structure.
  6. Select again the “Draw Style” tab.

The selection functionality is useful to show only part of a big molecule or to visualize different parts of the same structure with different representations.

The query string could be written directly or composed using the keywords and logical connectors offered.

Explore graphical rendering modes

  1. On the VMD Main window go to “Display > Rendermode” and change the graphical rendering mode to GLSL. If the graphic card supports it, atoms and bonds will appear smoother.
  2. Now toggle “Display > Depth Cueing”. Look how this mode changes the depth perceptions of the molecule. The effect is more visible with a molecule bigger than the one used for this tutorial.
  3. From the same menu toggle between Perspective and Orthographic projections. Perspective makes the scene more realistic but distorts proportions, so you do not know if something appears big because it is near or because it is really big.

All these graphical rendering controls are not here for aesthetic reasons, but to enhance depth perception and feature visibility.

Visualize orbitals

  1. From the VMD Main window select timestep 6 (Fig. 2) in the bottom-left numeric field at the left of the long slider.
  2. From the Graphical Representation window, push “Create Rep” and change the new representation to Orbital (see Fig. 5).
  3. Set its “Isovalue” to 0.05 and “Orbital” to 6.
  4. Change the “Coloring Method” to ColorID and the “ID” (the next menu) to 1 red to identify the positive lobe of the orbital.
  5. Create another Orbital representation pushing “Create Rep” and change its “Isovalue” to −0.05 and “Orbital” to 6.
  6. Change the “Coloring Method” to ColorID and the “ID” (the next menu) to 0 blue to identify the negative lobe.
  7. Close the Graphical Representations window.

This representation draws a molecular orbital isosurface corresponding to a user-defined wavefunction amplitude computed on a regularly spaced grid, resulting from the selected wavefunction type, spin, excitation, and orbital index. An isosurface is the surface that connects all points in a volume where the variable has a specific value set by the user. By convention the isosurfaces representing orbitals are draw at the ±0.05 level.

Export display image

  1. Open the File Render Controls window from the “File > Render…” menu on the VMD Main window.
  2. Select Snapshot from the “Render the current scene using” menu. With “Browse” select an output file name with extension .bmp and push “Start Rendering.”
  3. The generated image will show.
  4. Select Tachyon from the “Render the current scene using” and push “Start Rendering.” Tachyon is the VMD provided ray-tracer tool.
  5. The generated image will show. Note the differences with the one above (Fig. 9).
  6. Close the File Render Controls and the two image windows.

Export movie frames

  1. Go to “Extensions > Visualization > Movie Maker” on the VMD Main window (see Fig. 6).
  2. Push “Set Working Directory” and select an empty directory for which you have write permissions.
  3. Uncheck “Movie Settings > Delete image files.”
  4. Push “Make Movie.” Push “No” on the error dialog that says “Cannot locate videomach.exe” and “OK” to the subsequent Application Error dialog.
  5. The movie, as a sequence of images, is generated in the selected directory.
  6. Close the Movie Maker window.
Figure 6.

Access to the VMD extension plugins. The one shown here builds movies from a loaded trajectory. [Color figure can be viewed in the online issue, which is available at]

On Windows this extension requires the installation of a commercial tool to build the movie. But, as happens in the free software world, you can easily work around this requirement. The saved frames can be assembled into a movie file using “Windows Live Movie Maker” or “RAD Video Tools”[17] on Windows; mencoder[18] or ffmpeg[19] on any platform. More info on this procedure can be found on my visualization postprocessing page.[20]

Access extended capabilities

  1. Browse through the installed VMD extensions collected in the “Extensions” menu on the VMD Main window.
  2. Look to all VMD extensions available through the “Help > Script Library” and “Help > Plugin Library” menus.
  3. Launch the “Extensions > Tk Console” extension. It gives access to the VMD scripting language (see Fig. 7).
  4. Enter molinfo top get numorbitals. This command shows the number of orbitals present.
  5. Enter molinfo top get orbenergies. The result is a list of orbital energies.
  6. Enter molinfo top get orboccupancies to shows the orbital occupancy. Note that from this list you can determine the orbital number of the HOMO (orbital 4) and LUMO (orbital 5).
  7. Close the Tk Console window.
Figure 7.

The Tk Console extension gives access to the VMD scripting language. [Color figure can be viewed in the online issue, which is available at]

The VMD scripting language is Tcl [21] augmented with VMD-specific commands. For example from the VMD help (accessed through the “Help” menu) you learn that the molinfo command is used to get information about a molecule including the number of loaded atoms, the filename, the graphics selections, and so on. The top appearing in the commands above identifies a specific molecule which is used to determine some parameters, such as the center of view, the data in the animation controls, and so on. The top molecule is marked with a ‘T’ in the list of loaded molecules in the VMD Main window.

Save your work and exit

  1. To save the current status of your visualization, that is, the file loaded, the colors, the viewpoint. and other information, go to the “File > Save Visualization State…” menu on the VMD Main window.
  2. Enter a filename. By convention the status file has extension .vmd.
  3. Exit VMD with “File > Quit.”
  4. Restart VMD. To return exactly to the same state as before, reload the saved status file using the “File > Load Visualization State…” menu.
  5. Exit VMD with “File > Quit.”

Here this quick tour ends. To expand your VMD knowledge, besides the VMD User Guide, there are various tutorials on its site.[14] An interesting one is: “VMD Quantum Chemistry Visualization Tutorial.”[22] There are also various biomolecules under the “proteins” subdirectory of the VMD installation useful to explore more complex structures. Now we return back to a more general view of the visualization process.

Placing Visualization in the Discovery Cycle

The visualization process model shown in Figure 8 (adapted from Tory and Möller[23]) describes how visualization can be integrated effectively into the scientific discovery cycle. Inside this cycle visualization acts only as an interface, that is, as an adapter of numbers to human perceptual capabilities.

Figure 8.

Sketch of the visualization process embedded in the more general discovery cycle. [Color figure can be viewed in the online issue, which is available at]

Everything starts with a scientist who is studying something or has found something that piqued her curiosity. Surely she has in mind an initial model of this “something” that guides her data acquisition and initial visualization building. When a representation of the data is on screen, she interacts with it to better understand the object under study. Often she changes colors, visualization techniques, rendering options or viewpoint because: “A graphic is not ‘drawn’ once and for all; it is ‘constructed’ and reconstructed until it reveals all the relationships constituted by the interplay of the data.”[24] Simultaneously to data understanding, this exploration refines her mental model. In turn, this updated mental model influences her choice of visualization algorithms and graphical representations in a sort of feedback loop.[23]

Even in this seemingly straightforward process, the initial model the scientist has in mind not only guides, but can also prejudice the choice of visualization algorithms and bias the results interpretation because what we perceive is filtered by what we think to know. Even the tool itself could introduce artificial limitations because its available and highlighted functionalities often dictate the contribution it makes to the discovery process. In many cases, it seems the discovery process adapts to the tool functionalities and not vice versa as it should be.

Exploration and observation to elicit new questions

In the vast majority of cases, the cycle depicted above starts with a hypothesis that the scientist tries to prove or disprove. This is the classical, confirmatory data analysis cycle. Visualization helps here, but to produce indisputable results the scientist knows that should rely on quantitative analysis methods. Instead, what if the scientist has no idea, or hypothesis, over the object under study? Who suggests the right questions? Tukey affirms: “And roughly the only mechanism for suggesting questions is exploratory.”[25] Here visualization fully shows its usefulness by supporting an exploration-driven, evolutionary way to look at the data that could suggest its hidden model from which hypothesis could be drawn and scientific questions formulated.


With visualization the scientist's intuition and insight return to the center of the discovery cycle from where often had been displaced by calculation and simulation wonders. “The process of scientific discovery, however, is essentially one of error recovery and consequent insight. The most exciting potential of wide-spread availability of visualization tools is not the entrancing movies produced, but the insight gained and the mistakes understood by spotting visual anomalies while computing. Visualization will put the scientist into the computing loop and change the way science is done.”[26]

Coupling visualization and analysis

Visualization is not a substitute for analysis. With beautiful images there is always the risk of seeing regularities and correlations that are not there. “Humans are good… at discerning subtle patterns that are really there, but equally so at imagining them when they are altogether absent.”[27] Do not blindly accept what the images produced by a visualization tool suggest but always switch on your chemistry intuition.

Conversely, as remarks Richard Weinberg: “Computing, and in particular supercomputing, without visualization, is like assembling a jigsaw puzzle in the dark. If you can not see what you are doing, you are very likely to wind up with the wrong picture.”[28] That is, the wrong mental model of the phenomena under study. To be fully convinced, look at the famous “Anscombe's Quartet”[29] composed by four datasets that have nearly identical simple statistical properties, yet appear very different when graphed.

Results communication

“Discover the unexpected, describe and explain the expected” was the motto of the National Visualization and Analytics Center at the Pacific Northwest National Laboratory. It aptly describes the double role of visualization. One role supports exploration and cognition, in the other visualization creates beautiful images for results dissemination and communication.

Goal of visualization for communication is to present already comprehended results, usually through static (paper) or non-interactive media (movie), to people that should be persuaded, amused or called to action. Clearly this goal can be achieved only after visualization played its primary role as a discovery tool.

Teaching is a form of communication too and as such it could benefit from visualization. Instead of pouring abstract facts in the head of the students, the teacher could transform them into an interactive display, like has been done in ChemTube3D[30] where each web page, instead of showing an image of a chemical structure, had been enhanced and made interactive using a Jmol[31] 3D chemical structure viewer applet.

Equip Yourself

Visualization tool

The most obvious and immediate choice is to select a visualization tool already in use by your colleagues to capitalize on cooperation and mutual help. Failing this, you could look at Table 1 for a short list of suggestions.

Table 1. Initial list of free visualization tools for Quantum Chemistry
ToolWeb pageNote
  1. See also

Avogadrohttp://avogadro.openmolecules.netAlso editor
Vega ZZ 
YASARA Viewhttp://yasara.orgPartial support

Whatever your choice is, you have to answer two questions upfront: does the tool support the file formats your computations produce? A simple criterion is to select tools, like the ones suggested, that read and visualize at least GAMESS or GAUSSIAN log files, because these are the most common output formats from QM computations. The second question concerns whether or not you want to adopt a commercial tool. There are indeed good ones and they often are tightly integrated with the company's own computational codes. However, from my experience it is better to select, if possible, free, multiplatform tools to leave open as much alternatives as possible in the event of a workplace or computer change. Furthermore, being the tools free, you can experiment with several of them till you find the one that covers your needs. Whatever the choice is, it is a good move to verify upfront if the tool is actively developed and has a live community behind it.

Always keep in mind that no tool is perfect. The perfect tool makes everyone happy, but chemistry is a very wide design space and there is no single definition of perfection. It is better to start doing some visualization and learn “on the job” what the tool can do and how it could enhance the effectiveness of your research.

Machine hardware

Today it is really hard to find a computer not able to run a visualization tool. You only need to check that it is well equipped in term of memory and with 3D accelerated graphic card. Failing this, the visualization performance suffers and the responsive manipulation of the graphical objects becomes impossible.

Mono or stereo display?

One standard display, better if large, is all that is needed for visualization. But you could argue that we can put to better use our binocular vision to enhance depth perception using a stereo monitor. That is true and is even well supported by visualization tools.

However on the other hand, stereo does not avoid 3D scene intrinsic perception problems, like mutual occlusion between its various parts or failure to perceive dimensions and relative positions. It is much better to stay with a non-stereo monitor and rely on interactive manipulation of the viewpoint to build the correct mental model of the visualized data.

Learning Visualization

“Drawing graphs, like motor-car driving and love-making, is one of those activities which almost every [researcher] thinks he can do well without instruction. The results are of course usually abominable.”[32] This statement depicts what sooner or later anyone has seen projected on the screen at a conference.

To avoid falling in this trap, before firing any tool, spend some time thinking about what visualization means. Peruse resources like the “Data Representation in Chemistry”[33] page. Look around for examples in your and other fields' publications and conferences. Look actively at these images asking yourself why they work or not and checking if they really force you to see what you were not expecting to see.

To learn the technicalities of a visualization tool, nothing substitutes direct experience. Try a visualization tool yourself, better if you have at hand a concrete problem to solve. In this phase I usually suggest, if possible and if available, to collaborate with a visualization expert to use these tools efficiently and to have two pairs of eyes looking at the problem rather than spending too much time with new technology. Only after that, it is effective to take one of the various tool's tutorials available.

Visualization Side Fields

Data management

We often put more effort to maintain our collection of music files than to manage our computational results. Nonetheless data management for visualization is extremely important, otherwise we risk gaining meaningless insights from computational results. This happens, for example, when we draw conclusions from a visualization thinking it derives from a different set of parameters, or when we are unable to reproduce a given published image from a set of unnamed data files. You see, data management for visualization does not concern only with file formats and associated readers.

Regrettably current tools seldom help in managing data and common file formats provide scarce support to record additional information. So are the researchers who should devise methods to record data provenance and meaning of their data.

Visual communication

Every visualization tool is able to export images or build movies from what it shows. But to turn those outputs into real communication you should pay attention to the expected audience, to what you want to transmit and to the final goal of the communication. Otherwise a confused hardcopy of the screen will convey only the confusion reigning in your mind. Remember that the goal of communication is seldom to simply transmit information. The goal is almost always to convince the audience of the importance of your results. To enhance visual communication effectiveness, there are plenty of resources covering these aspects, for example, the “Practical Visualization for Chemistry Course” slides.[34] Here below, I focus only on few easily forgotten technical issues.

Image Formats

Usually visualization tools can save images in a variety of formats. On the other hand, specific image uses (web page, presentation, publication) each require a suitable image format. Normally there is no problem in converting from one set to the other using image manipulation tools. However, I suggest checking carefully the produced images for artifacts generated by their sharp edges and hard contrasts before committing to a specific format. On my side, I found PNG format adequate for all uses.

To have nice pictures on a printed page, the original image should have a high number of pixels. A rule of thumb states that for printing the image should have six times the side pixel count compared to a similar sized image on screen. Visualization tools, when taking a snapshot of the current displayed scene, produce output images as big as its viewer window (this does VMD using the “File > Render > Snapshot” option). To overcome this limit, tools like VMD can export the graphical scene to a ray-tracer that produce professional looking images without size limits (see Fig. 9).

As you can see, besides the visualization tool, to publish nice and meaningful images a toolbox of image manipulation and conversion programs is often needed together with a collection of related tricks, like the ones found on the “Post processing and visual communication” page.[20]


Visualization uses colors to identify or distinguish things (e.g., color by atom type, line color in a multiline chart), to code quantitative values (e.g., electron density values) through a colormap (see Fig. 4) and to draw attention to specific points on the graphical structures.[35]

In these multiple roles, color is an important tool for cognition amplification. Thus color choices should be carefully considered without automatically relying or trusting tool's defaults. A case in point is the “rainbow colormap” offered as default by almost all visualization tools even if research has shown that it is rarely the optimal choice when displaying quantitative data.[36]

Remember also that the visualized image appears different on media and devices other than the computer screen.[37] Colors that are vibrant on screen could be almost invisible on a conference room beamer (usually the green is barely visible) or indistinguishable on a printed page where available colors only partially overlap with those visible on the computer monitor. The situation becomes even worse if you have to publish on a journal that still requires black and white images.

There is no quick recipe to select the best colors. Always experiment and visually check if you can distinguish colored objects on the printed page or on the conference hall's projection screen.[38]


Making a movie out of a visualized trajectory seems the natural thing to do. But these movies often suffer from two diseases: being too long and the “Just One Darned Thing After Another” syndrome. In both the cases, the significance of the data is rapidly lost.

Instead, before creating a movie, clarify the focus of the presentation and think about what you want to show. Maybe shortening it and adding stop frames and identifying elements, like arrows, to draw attention to the right spot is sufficient to make the message pass. In extreme cases, avoid movies altogether and simply show significant frames of the trajectory.

Sounds like a trivial suggestion, but always check if your movie plays on the machine where you decided to show it. It is quite common that the movie plays on your office machine but refuses to show on the conference hall one.

Multitechnique and Multidata Composition

Often a visualization, to be effective, should combine more than one visualization technique or visualizations of different data. These compositions create an interpretative context for the main data, suggest correlations or make clear cause-effect relationships.

For example Figure 10 presents the results of a study on LiAl alloys.[39] This image combines the Al lattice visualization with electron concentration along its bonds, the Li atoms and the electron density contours along a specific plane to make visible the main results of the study.

Figure 9.

Structure and orbitals rendered as screen snapshot (left), where they appear exactly identical to the screen, and exported using a ray-tracer tool (right). Note the shadows and the more realistic reflections on the surfaces. The visualized data is GAMESS EXAM36.LOG MCSCF Optimized orbital 7. [Color figure can be viewed in the online issue, which is available at]

Figure 10.

Visualization of a LiAl lattice and the corresponding electron density represented as isosurface and as contours on a cutting plane. (Data from C. Cucinotta, unpublished, used with permission). [Color figure can be viewed in the online issue, which is available at]

Future Challenges

Big data

Usually QM results are relatively small, but are becoming more and more common huge datasets resulting, for example, from long timescales needed to simulate many important biological processes. Even storing trajectory frames infrequently, the resulting petabytes of data to analyze will be far too large to move, so visualization and analysis will be done primarily on the supercomputer where the data already resides.

Currently, this requires bringing the visualization tool display back to the desktop computer screen by one of the various methods available. For example, by relying on a remote desktop, like TightVNC[40] or on a graphics accelerated remote display like VirtualGL.[41]

The trend toward increasing data sizes suggests having visualization tools integrated with computational codes, reading data directly from their memory structures. This way no data should be stored and retrieved and, more important, the researcher acquires the feeling of how the simulation is evolving. This is called in situ visualization for which early projects, for example with VMD,[42-44] are underway.

Big data mean also graphical representations should change. It makes no sense to show million-atom biomolecule using ball and stick structural representation. There are too many details that distract and do not contribute to data comprehension. Thankfully representations have been invented just to overcome these problems. For example, VMD offers the “Beads” representation that shows bounding spheres for proteins' residues reducing the number of graphical primitives drawn, but still giving a crude space-filling overview of the structure.

Expanding representations

Visualization is a highly conservative field: tools developers often reinvent and reimplement the same functionalities over and over using the same graphical representations. For their part, users think the representations and techniques implemented are the only ones possible. Consider that: “Working with a stagnant set of representations creates a more subtle and profound problem. By restricting the set of available visualizations, we limit the ways that scientists think about their models and thereby limit potential insights.”[45]

Only tighter interaction between users and visualization tools' developers will overcome these artificial limitations. Ask yourself if you could suggest a new representation of data or a different visualization method that could push forward your research. In these contributions, you are free to invent new and useful representations because the chemistry ones are metaphorical and are not constrained by real-world geometries.

Representations could be expanded also by incorporating visualization techniques that are widely used by other scientific fields. Examples are the use of illuminated streamlines to visualize ring currents in the presence of an external magnetic field[46] or the integration of fluid dynamics specific visualization techniques inside an atomistic visualization.[47]

Graphic cards already provide enhanced rendering, like advanced surface textures or ambient occlusion to simulate realistic lighting. Incorporating these capabilities into the visualization tools will enhance visualization understanding by providing fluid interaction with the graphical scene and by offering more perceptual cues.[48, 49]

Reproducible research

If visualization is regarded as an important step in data understanding, then becomes crucial to save its artifacts and the process that lead to them. That is, making visualization reproducible means to tie specific instructions to data analysis and experimental data so that scholarship can be recreated, better understood, and verified.

Tool support is needed to make visualization reproducible. Besides the current visualization status recording functionalities, what will be needed is a way to preserve crucial provenance information too: how the input files have been generated, using which parameters, if preprocessing steps had been applied and so on.

Away from the desktop

As already happened in other fields, the mobile revolution is infiltrating also science. But simply porting a visualization tool to the smartphone, as has been done, for example, with the Jmol Molecular Visualization App,[50] is not enough. New ideas will be needed to explore what chemistry visualization could become on smartphones and tablets.


Chemistry is an eminently visual science and chemists are visual people. Thus, it is natural to predict a widespread adoption of visualization in quantum chemistry. Nonetheless, it is time to answer few questions.

Is visualization useful? Definitely yes. It is an effective tool to meet the challenge of new, more complex data interpretation tasks.

Should I care about visualization in my research? Yes. Visualization has proven to work, so it is a pity not to give it a chance to help you.

Should I invest time to learn visualization? To learn how to produce effective visualizations, yes. However, more than on books, you learn how to produce useful visualizations actively looking around, studying existing examples and trying them in your work.

Is there any roadblock on the path to a more generalized use of visualization in quantum chemistry? Yes, the obstacle is considering visualization a way to produce nice images only and not a data understanding process.

However, when considering how to approach visualization, never forget that its final goal is to use vision to amplify cognition. “Using Vision to Think.”[2] That is.


The author thanks C. Cucinotta for allowing him to use her study's image, and also thank the CSCS management for supporting this work and the reviewers for their helpful comments.


  • Image of creator

    Ing. Mario Valle works since 2003 at the Swiss National Supercomputing Centre (CSCS) helping researchers and scientists to extract the meaning hidden in the computational results produced by the CSCS' supercomputers through data analysis and visualization. Help and cooperation that led him to take part in studies and research in many different fields whose results reported and shared in about 30 publications and books chapters.

    Prior to joining CSCS he was with Advanced Visual Systems (AVS) defining and implementing innovative visualization projects for the company's customers and with Digital Equipment (DEC) working as software developer and software engineering consultant.

    Ing. Mario Valle holds an Electronic Engineering Degree from Università di Roma “La Sapienza” and is IEEE Computer Society member. [Color figure can be viewed in the online issue, which is available at]